In the fault diagnosis of hybrid systems, the uncertainty of the system and the interference of the outside world lead to a large number of false positives and false alarms in the diagnosis process. To solve the above problems, a global analytical redundancy relationships (GARRs) is obtained by using the hybrid diagnostic bond graph (HDBG) model. according to the GARRs, the uncertainties of system parameters under normal working mode, gets the adaptive threshold. Combining the adaptive threshold with the fuzzy method, it realizes the fault detection and isolation (FDI) of parametric faults in uncertain hybrid systems. In this paper, the flying capacitive three-level inverters are studied and simulated in 20-sim software. The simulation results verify the feasibility and superiority of this method.