Real-time junction temperature estimation of IGBT is particularly important to ensure the reliable operation and health management of converters for energy storage. In this paper, an adaptive Kalman filter (AKF) is used to combine the temperature-sensitive electric parameter (TSEP) method with thermal network model to optimize the noise in the (TSEP) measurement in real time and overcome the problem that the thermal network model is susceptible to the influence of the boundary conditions. The excellent performance of junction temperature estimator based on AKF is verified through several case comparisons.