Data Reduction for Dynamic Stability Classification in Power System

Abstract

A large oscillation of power system caused by faults must be quickly treated so that the opportunity driving power system into re-stability state can be easier. Necessity is to select a compact data-set that is the representative of all data-sets with the aim of reducing computing costs, reducing computer memory, and improving classification accuracy. The K-means (KM) algorithm is the most commonly used clustering algorithm because it can be easily implemented and is the most effective one in terms of the execution time on large data size. The key problem of KM is that it is sensitive to initial center and may converge to a local optimization. In this paper, we proposed the use of Hybrid K-means (HKM) data clustering algorithm that can avoid being trapped in a local optimal solution. The HKM was applied with the aim of reducing data space. We also proposed a process data clustering applied to classify the problem of dynamic stability in power system. The K-Nearest Neighbor (K-NN) was chosen as the classifier. The K-NN participated in the evaluating classification accuracy stage. The study was done on IEEE 39-bus. The results showed that the proposed algorithm achieved effective data size reduction and high accuracy classification.