A study on temperature monitoring method for inverter IGBT based on memory recurrent neural network

The power module of the Insulated Gate Bipolar Transistor (IGBT) is the core component of the traction transmission system of high-speed trains. The module's junction temperature is a critical factor in determining device reliability. Existing temperature monitoring methods based on the electro-thermal coupling model have limitations, such as ignoring device interactions and high computational complexity. To address these issues, an analysis of the parameters influencing IGBT failure is conducted, and a temperature monitoring method based on the Macro-Micro Attention Long Short-Term Memory (MMALSTM) recursive neural network is proposed, which takes the forward voltage drop and collector current as features. Compared with the traditional electrical-thermal coupling model method, it requires fewer monitoring parameters and eliminates the complex loss calculation and equivalent thermal resistance network establishment process. The simulation model of a high-speed train traction system has been established to explore the accuracy and efficiency of MMALSTM-based prediction methods for IGBT power module junction temperature. The simulation outcomes, which deviate only 3.2% from the theoretical calculation results of the electric-thermal coupling model, confirm the reliability of this approach for predicting the temperature of IGBT power modules.