Construction of a Dynamic Monitoring and Early Warning System for Voltage Stability in Transparent Grids with High-Density Distributed Power Sources Based on Long Short-Term Memory Networks

Abstract

With the large-scale integration of high-density distributed generation (DG) into distribution grids, the randomness and volatility of their output pose severe challenges to grid voltage stability. Traditional static analysis methods based on physical models struggle to meet the demands of real-time dynamic monitoring and early warning. Leveraging the real-time and efficient acquisition of power data through transparent grids, this paper proposes a voltage stability assessment method for high-density DG integration into transparent grids. This method employs a graph convolutional network (GCN), a bidirectional long short-term memory (BiLSTM) network, and an attention mechanism. Case studies demonstrate that both line-carried power and the longitudinal component of voltage drop positively correlate with voltage stability, enabling rational DG power regulation. Furthermore, on the hybrid dataset D, the GCN-BiLSTM-Attention model achieves an accuracy of 96.62% and an F1 score of 98.38%, indicating high predictive precision. This method enables precise delineation of system instability zones, facilitating emergency control operations in power systems.