The phasor measurement unit(PMU)can directly obtain measurement data such as the rotor angle of generators in the dynamic process. However

considering random noises in the real-time measurement data

it is necessary to filter out the noises in the measurement data to get more accurate information on generator states. This paper presents a novel method based on the unscented particle filter(UPF)to dynamically estimate the states of synchronous generators. Firstly

dynamic state estimating models of the generators are developed based on the fourth-order dynamic equations. Secondly

in the framework of the particle filter(PF)

the proposed method obtains the important density function of PF by using unscented Kalman filter(UKF). The proposed method takes the latest measurement information into consideration in the process of generating predictive particles

which makes the distribution of particles much closer to the posterior probability distribution of the true states. Finally

the performance of the proposed method is compared with UKF and PF both in American Western Systems Coordinating Council(WSCC)3-machine 9-bus system and an actual grid. Simulation results show that

compared with PF and UKF

the UPF performs better in estimation precision and robustness to measurement noise.