A Data-Driven Method for Power System Transient Instability Mode Identification Based on Knowledge Discovery and XGBoost Algorithm

Abstract

Aiming at the difficulty of unstable pattern recognition after power system fault, a novel identification framework for transient instability mode identification based on knowledge discovery by accuracy maximization (KODAMA) and extreme gradient boosting (XGBoost) algorithm is proposed. In this method, the transient stability of all typical fault scenarios of power system is obtained firstly by XGBoost. Then, to make full use of the structure of the raw data and mine the contained data information, a novel data mining algorithm KODAMA is introduced to cluster the mode of rotor angle in case of instability, thus to convert pattern-unlabeled case data into pattern-labeled data. Finally, based on this labeled data, to fully reflect the dynamic characteristics, a multiple XGBoost assessment strategy is designed to recognize different instable modes. The proposed technique is tested on the Nordic test system, and the results indicate that the proposed approach can provide fast and accurate recognition of instable mode and has a certain prospect of online application.

Keywords

• Transient stability
• KODAMA algorithm
• XGBoost algorithm
• machine learning