Reducing computational complexity in nonlinear power system state estimation via ANN-assisted linear Kalman filtering
Abstract
This paper proposes a computationally efficient dynamic state estimation algorithm for nonlinear power networks using synchronized phasor measurements and a combination of an artificial neural network and a linear Kalman filter. The methodology involves using both the predicting abilities of the artificial neural network (ANN) and the iterative correction properties of the linear Kalman filter to estimate system states under dynamic operating conditions. Instead of using the nonlinear estimation approaches requiring the repeated computation of Jacobians and matrices, the presented approach significantly reduces the computational complexity while maintaining accurate estimation. The ANN is first trained on PMU measurement data and generates predictive state values for the system, which are corrected using the linear Kalman filter to remove the effect of noise and increase resistance to measurement corruption and malicious attacks. Numerical simulation experiments carried out on the IEEE 6-bus test network have shown that the hybrid state estimator provides more accurate results in terms of reduced RMSE and faster convergence when compared to the conventional state estimation methods.