Neural Network-Based Dynamic State Estimation for Fast Frequency Support Using Energy Storage Systems

Due to their lower inertia, microgrids are susceptible to rapid frequency changes and excursions. Energy storage systems (ESSs) can provide fast-frequency support (FFS) to mitigate these issues. However, providing FFS requires measurements of fast frequency dynamics which is challenging to achieve with noisy measurements as the use of traditional low-pass filters creates unwanted delays. While classical approaches like moving horizon estimation (MHE) address this problem, they rely upon knowledge of noise statistics, which is often not available, and is computationally expensive. In this paper, we propose a neural network-based dynamic state estimator for estimating the frequency dynamics of microgrids, trained with observed data, and adaptable even when the system model is unknown. (In this case, system dynamics is also identified as a byproduct.) The performance of the proposed state estimator is compared with that of MHE through MATLAB/Simulink simulations, demonstrating comparable accuracy with superior computational efficiency.