The present invention discloses a real-time online prediction method for a dynamic junction temperature of a semiconductor power device. The present invention has advantages as follows: the sampling value of electrical parameters required for system closed-loop control is multiplexed as inputs, and no additional system hardware circuits and costs are needed; the processor resources can be saved to the utmost extent by using the idea of discrete iterative calculation, online calculation can be realized, and real-time performance of dynamic junction temperature calculation can be ensured; an optimal fitting dynamic thermal resistance discretization model is creatively proposed, and is used to perform iterative calculation, so that while real-time performance of dynamic junction temperature calculation of the power device is ensured, calculation accuracy is also ensured, meeting the requirements of protection, life prediction, and reliability design of the power device, and this method is very suitable for actual engineering application.