Aiming at the problem of DC side short-circuit fault diagnosis of flexible DC system, this paper conducts research from accurate modeling of fault network and analysis of transient characteristic signals, and proposes a fault diagnosis method based on the correlation of time-frequency characteristics of impedance network transient response. The method injects a pulsed voltage into the DC system line entrance as a stimulus, extracts the transient current component generated in real time, and obtains the time-frequency energy sequence through different scales of the Mexican Hat transform. And the energy spectrum matrix which can reflect the time-frequency characteristics of the transient response of the fault network is constructed. At the same time, an accurate model of a flexible DC system is established to simulate short-circuit faults with different resistances at different fault points, and a complete fault sample and its time-frequency characteristic database are constructed. The similarity calculation can select the closest sample to the measured fault characteristics to determine the fault Point location and estimated short-circuit resistance. Finally, an example analysis of the above-mentioned principle and method is performed, and the comparison with the time-domain method shows that the diagnostic method and its criteria described in the article are more accurate and effective, and the relevant principles are correct and feasible.