Real-Time Detection of Power System Disturbances Based on $k$ -Nearest Neighbor Analysis

Abstract

Efficient disturbance detection is important for power system security and stability. In this paper, a new detection method is proposed based on a time series analysis technique known as ${k}$ -nearest neighbor ( ${k}$ NN) analysis. Advantages of this method are that it can deal with the electrical measurements with oscillatory trends and can be implemented in real time. The method consists of two stages, which are the off-line modeling and the on-line detection. The off-line stage calculates a sequence of anomaly index values using ${k}$ NN on the historical ambient data and then determines the detection threshold. Afterward, the on-line stage calculates the anomaly index value of presently measured data by readopting ${k}$ NN and compares it with the established threshold for detecting disturbances. To meet the real-time requirement, strategies for recursively calculating the distance metrics of ${k}$ NN and for rapidly picking out the ${k}$ th smallest metric are built. Case studies conducted on simulation data from the reduced equivalent model of the Great Britain power system and measurements from an actual power system in Europe demonstrate the effectiveness of the proposed method.