A real-time state estimator based on WAMS is presented， in which the recursive extended least-squares algorithm is used to identify the state transition matrix， and the fading memory exponential weighted algorithm is employed to estimate the error covariance matrix of the linearized model. At the same time， the measurement weights are calculated on-line by the measurement standard deviations. When the filter is done by using observed data， the imprecise（or unknown） parameters and the noise covariance matrixes of the state estimation model are constantly identified and corrected on-line simultaneously， which makes the model follow the contingencies and the time-varying process adaptively. Therefore the estimated errors are reduced， and the filtering precision is high. Simulation results show that the proposed method has good performance under different scenarios such as normal operation， bad measurements， sudden load change/drastic generation variation and topology errors.