The junction temperature measurement of insulated gate bipolar transistor based on multi-layer feed -forward neural network is presented

Abstract

For large power flow type insulated gate bipolar transistor (IGBT) saturation voltage drop and non-linear relation between the collector current and junction temperature of junction temperature prediction problem, set up a big electric saturated pressure drop down IGBT test system, access to settle moderate collector current and saturation nonlinear relation between the pressure drop curve, analyses the relation curve change rule the corresponding physical mechanism. The error back propagation (BP) neural network model was established by using Matlab software to predict the junction temperature. The comparison with the prediction results of the polynomial mathematical model shows that the relative error and prediction error of the neural network model are smaller than that of the polynomial mathematical model, and the prediction accuracy of the junction temperature is higher. It shows the feasibility of neural network for junction temperature prediction, so BP neural network model can be given priority when selecting junction temperature prediction model.