Power MOSFET Lifetime Prediction Method Based on Optimized Long Short-Term Memory Neural Network

As the core of conventional power electronics, the reliability problem of Metal-Oxide-Semiconductor Field-Effect Transistors (MOSFETs) severely restricts the safe operation of the equipment. Accurate prediction of the remaining useful life (RUL) of MOSFETs is the key to achieve prognostic and health management (PHM) and condition-based maintenance (CBM). In this paper, long short-term memory (LSTM) networks are combined with adaptive moment estimation algorithm, Dropout techniques and Bayesian optimization methods to improve prediction accuracy and generalization by optimizing model parameters with continuously updated probability distributions. The results show that compared with exponential fitting and traditional LSTM methods, the method has the advantages of small prediction error, high prediction accuracy and good prediction stability, which is beneficial to practical engineering applications.