The invention discloses a dynamic state estimation method based on a self-adaptive EnKF technology, which is used for dynamic state estimation of a generator of a power system under an unknown systemnoise working condition. According to the invention, an improved fading memory index weighted Sage-Husa noise statistical estimator is introduced into the EnKF, which can dynamically estimate and correct the mean value and the variance of time-varying system noise and inhibit the influence of unknown system noise on the state estimation precision, so that the accurate estimation of the generator operation state is achieved. Due to the fact that the actual engineering background is considered, the method is simple and convenient and has high engineering application value.