Neural Network Applications in Hybrid Data-Model Driven Dynamic Frequency Trajectory Prediction for Weak-Damping Power Systems
Guoteng Wang, Chongyu Wang, Mohammad Shahidehpour
Abstract
This article proposes a hybrid data-model driven dynamic frequency trajectory (DFT) prediction method for achieving an accurate perception of low-inertia and weak-damping power systems. First, an improved primary frequency regulation (PFR) model is established for calculating the dominant oscillation mode. Different from the existing PFR models, the proposed PFR model has fully considered the spatial frequency distribution of weak-damping power systems. Then, a novel neural network topology based on edge aggregated graph attention networks (EGAT) and long short-term memory (LSTM) is proposed for the online prediction of maximum and steady-state frequency deviations, by which the temporal features of frequency can be accurately extracted. In addition, multiple EGAT-LSTM units with various input time windows of different lengths are cascaded to boost the proposed prediction accuracy. Next, a hybrid data-model driven DFT prediction method is proposed, which is based on the improved PFR model and the proposed EGAT-LSTM model. Finally, the proposed method is verified on the IEEE 39-bus system. The numerical results demonstrate that the proposed method can accurately predict the DFT, which outperforms the existing solution methods.