On-line learning applied to power system transient stability prediction

Abstract

A neural network-based system is proposed for power system transient stability prediction. A power system is a nonstationary environment, where operating conditions change from time to time. To make accurate predictions of the transient stability status of a power system, training examples are added continuously to reflect the most current operating condition. An on-line learning algorithm is employed to accommodate new training examples while avoiding negative interference. A real-world power system in China is used to demonstrate the effectiveness of the proposed transient stability prediction system. Simulation results show that the system performs well in different working modes and is able to make accurate predictions.