In order to accurately detect the time-frequency characteristics of complex non-stationary disturbance signals in the power grid and improve the time-frequency localization ability of the Hilbert - Huang Transform (HHT), This paper presents a method to detect power quality disturbance based on improved HHT. For voltage drop and short time interrupt disturbance signal, harmonic and compound disturbance signals, the method adopts the moving average method to average the instantaneous amplitude frequency parameters obtained by HHT, then extracts the energy density of the signal in different time and frequency from the Hilbert spectrum, locates the start-stop time of disturbance signals. Simulation results show that the method can obtain the frequency component, amplitude and mutation time of harmonic signal accurately and quickly, analyze the amplitude and start-stop time of voltage drop and short time interrupt signal. It is also suitable for the detection of compound disturbance signals, has higher precision and time-frequency resolution compared with the traditional HHT method.