The invention relates to a mechanical fault diagnosis method of a gas insulated switchgear based on abnormal vibration. The method comprises the steps of taking a vibration signal of a GIS in a normalstate as a reference, judging that a vibration signal is abnormal, and then carrying out the feature extraction; adopting the wavelet packet decomposition to process the abnormal vibration signal, obtaining the wavelet packet feature entropies of all frequency bands, and constructing the feature entropy vectors; and constructing a data set according to the feature entropy vectors, and performingfault mode identification by adopting a BP neural network. According to the invention, by utilizing the advantage that the wavelet packet energy entropy has multiple resolutions, the time-frequency localization depiction advantage is obvious, and the method is very suitable for processing the stable and non-stable vibration signals; meanwhile, by overall planning the vibration signals under the action of four excitation forces including a circuit breaker, an isolation switch, a grounding switch and an electromagnetic force, the data information is comprehensive, the diagnosis reliability is high, and a calculation result is accurate.