The invention discloses a method based on an improved HHT algorithm and applied to time-frequency analysis of GIS (Gas Insulated Switchgear) mechanical vibration signals. The method comprises the following steps: measuring GIS normal operation vibration signals and GIS vibration signals under faults, extracting characteristic quantities of the vibration signals by using ensemble empirical mode decomposition (EEMD), and carrying out hilbert transformation (HT) on the signals obtained by EEMD preprocessing to carry out time-frequency analysis on the GIS mechanical vibration signals. Through the method, the GIS vibration signals can be effectively processed by carrying out time-frequency analysis of the GIS mechanical vibration signals through adoption of a method of carrying out HT after preprocessing via an EEMD algorithm, so that a GIS mechanical fault diagnosis database is established; the theoretical basis is provided for achieving on-site live detection of the GIS mechanical faults.