Contingency-based Voltage Stability Monitoring via Neural Network with Multi-level Feature Fusion

Abstract To monitor the voltage stability state of complex power grid, a four-category stability classification problem that incorporates a set of serious contingencies is posed. Quick decision-making and high accuracy are critical for the safety operation of power system. However, this problem involves feature of different types, levels and dimensions and is hard to be handled by the traditional classifier. This paper utilizes the deep learning technique and proposes a multi-level deep neural network (ML-DNN) that achieves feature fusion of the electrical parameter measurements, topology and contingency information. experiments are implemented on IEEE-39 system, the ML-DNN performs better in four main evaluation indices comparing with five existing models, which demonstrates its advantage for online voltage stability monitoring.