IGBT Junction Temperature Prediction Model based on GA-LM-BP Algorithm

Abstract

Aiming at the problem that the Insulated-Gate Bipolar Transistor (IGBT) chip is encapsulated inside the module and the chip junction temperature cannot be measured directly, an IGBT junction temperature prediction algorithm model based on the back-propagation neural network algorithm optimised jointly by the genetic algorithm and the Levenberg-Marquardt (LM) algorithm is proposed. Firstly, the temperature-sensitive electric parameter (TSEP) method is used to build an experimental platform for the saturation voltage drop of IGBT modules; then, 2250 sets of saturation voltage drop, collector current, and IGBT case temperature data are extracted from the experimental data as feature inputs to characterize their relationship with the junction temperature of the IGBT module; finally, a junction temperature prediction model is established by using the extracted electrical parameters in the Genetic Algorithm-Levenberg-Marquardt-Back Propagation (GA-LM-BP) neural network algorithm to predict the junction temperature. Finally, the GA-LM-BP neural network algorithm is used to establish a junction temperature prediction model by using the extracted electrical parameters. The experimental results show that the average absolute percentage error of the junction temperature prediction value of the GA-LM-BP neural network algorithm is 0.114 when the collector current is less than the critical current, and 0.062 when the collector current is greater than the critical current, which is more accurate than that of the GA-optimised BP neural network algorithm and classical BP neural network algorithm in predicting the junction temperature of the IGBT module.