Power systems
AmirHossein Babaali; Mohammad Taghi Ameli
Abstract
Short-term voltage stability (STVS) varies with operating conditions of power networks, making its accurate assessment a critical challenge. This paper investigates a multi-class, data-driven approach to STVS evaluation. A dynamic index is employed to categorize voltage magnitude variations into three ...
Read More
Short-term voltage stability (STVS) varies with operating conditions of power networks, making its accurate assessment a critical challenge. This paper investigates a multi-class, data-driven approach to STVS evaluation. A dynamic index is employed to categorize voltage magnitude variations into three classes: stable, alert, and unstable. A significant obstacle in data-driven methods is missing measurement data, typically caused by sensor failures or communication delays. To address this issue, we propose two complementary solutions. First, a Bidirectional Gated Recurrent Unit (Bi-GRU) network with an attention mechanism is designed to recover data loss due to sensor failures. This method leverages both temporal trends and historical system information to reconstruct missing values with high accuracy. Second, a variable-length sliding window (VLSW) algorithm combined with a Bi-GRU is introduced to mitigate data loss arising from communication delays. The VLSW algorithm enhances data diversity and enables fast recovery. Simulation results on IEEE 39-bus and IEEE 118-bus test systems demonstrate that the proposed framework effectively identifies multi-class STVS under missing data conditions and remains robust against long-range data losses. Finally, validation on a real-world local network further confirms the practicality and robustness of the proposed approach.
