The invention discloses a method for classifying power quality mixed disturbances based on time-frequency domain multi-feature quantities. Classify voltage sags, voltage swells, voltage short-time interruptions, pulse transients, oscillation transients, harmonics and flicker power quality disturbances and their combined disturbances. The specific implementation steps first use clustering The empirical model decomposition method EEMD and the improved incomplete S-transform MIST process the disturbance signal, and extract 9 time-frequency domain feature values; then, the feature quantities are input into the block-based automatic classification system for disturbance identification. This method fully considers the mutual interference between single disturbances, and effectively suppresses them through complementary time-frequency domain feature quantities. The simulation results show that under certain noise conditions, the method can effectively classify power quality disturbances such as voltage sags, voltage swells, voltage short interruptions, pulse transients, oscillation transients, harmonics and flicker and their combinations. into a mixed disturbance.