A real-time diagnosis method, which is based on probabilistic Petri nets, has been proposed to solve the problem of inability to accurately locate the real-time fault source when the traction system reports a functional failure during train operation. By mining the dynamic temporal series change law between the fault source related to functional faults and operating events related to functional faults, a probabilistic Petri net model corresponding to various fault sources has been established, with the diagnosis decision to be made based on real-time calculation model output probability values, thus achieving a rapid and accurate location of functional faults. Based on on-site case data testing of inverter overcurrent faults, the proposed method can accurately locate six types of typical fault sources that lead to inverter overcurrent, with a diagnostic response time of less than 0.1 s. Compared with threshold detection and offline diagnosis methods, this method significantly improves the real-time diagnosis with its robustness under non-stationary conditions by dynamically adjusting weights and supeimrposing concurrent fault probabilities, thus providing an effective solution for traction transmission system functional faults as well as implementation of differentiated protection strategies.