Robust Dynamic State Estimation of Doubly-Fed Induction Generator Considering Measurement Correlation in Wind Farms

Wind power is characterized by strong intermittency, random fluctuating and low schedulability. Dynamic state estimation (DSE) of doubly-fed induction generators (DFIGs) based on phasor measurement unit (PMU) can provide reliable data basis for wind farm energy management system, which helps to suppress the influence of wind power fluctuation on power grid. However, DSE of a single DFIG needs input its terminal measurements, but currently it is not common for wind farms (WFs) configuring PMUs for each wind turbine. Besides, DSE based on the prediction-filtering framework of Kalman filter is susceptible to bad measurement data and disturbances. To address these issues, this paper proposes a robust DSE method of DFIG considering the measurement correlation in WFs. Firstly, considering that a WF can be considered as a small radial grid, its topological characteristics are leveraged to obtain the optimal installation locations of PMUs. And the time-dimension constraint information of DFIG dynamic model is included to construct the redundant measurement set. Secondly, the terminal electrical quantities of all DFIGs are estimated by weighted least absolute value (WLAV) estimation method based on the redundant measurement set. This way, the bad data in the DSE inputs and measurements can be filtered out by WLAV after distinguishing disturbances and bad measurement data. Finally, when the innovation of cubature Kalman filter is abnormal, it can be considered that the prediction vector deviates from the true value. The process noise covariance matrix is then adjusted by introducing a process noise scale factor, which can reduce the influence of the predicted values on the final estimation results. Thus, this integrated method has low requirements for measurement configuration and can deal with disturbances and bad data in the DSE inputs and measurements simultaneously. Simulation results on a WF with 16 DFIGs show that, only 5 PMUs are needed to make all buses of the WF observable, and the dimension of the measurement set increases from 34 to 50 by incorporating the predicted terminal currents of all DFIGs as virtual measurements. As the measurement redundancy increases, the root-mean-square error (RMSE) of the estimated DFIG terminal electrical quantities reduces from 0.251 to 0.009 which is close to the RMSE of the ideal measurement set. Besides, based on the spatial correlation of the PMU measurements at the different buses, the voltage drop disturbance at the point of common coupling and bad measurement data at Bus 6 are correctly detected and discriminated. And when bad measurement data and disturbances exist, the mean RMSE of the dynamic states estimated by the proposed method is lower than other comparison methods. For example, the RMSE of rotor speed drops from 0.009 to 0.003. Simulation is also carried out on an actual wind farm in a certain area to verify the performance of the proposed method for large scale wind farms, and the results demonstrate its good generalization ability. The following conclusions can be drawn from the simulation analysis: ① With the optimal configuration of PMUs and the spatial correlation of phasor measurements, all wind farm buses can be observed with a small number of PMUs. The temporal correlation of DFIG dynamic states can be leveraged to construct the redundant measurement to improve the estimation accuracy. ② The spatial correlation of PMU measurements can be used to effectively distinguish disturbances and bad measurement data. And the robust WLAV method is used to estimate the terminal electrical quantities of all DFIGs. It can also filter out the influence of PMU bad data on input vector and measurement vector of DSE, which is difficult to cope with during the DSE execution. ③ The process noise scale factor introduced to reduce the weight of prediction value is able to deal with the inaccurate predicted states of DFIG when a disturbance occurs. © 2023 Chinese Machine Press. All rights reserved.