Online short-term voltage stability monitoring based on wide-area trajectory data analytics

In this chapter, a TS data analytics-based imbalance learning machine for power system online SVS monitoring is systematically developed. It is dedicated to addressing the challenging class imbalance problem in practice, which is induced by the assumption that learning samples are mainly collected from historical and/or online PMU records. To deal with class skewness from both data-preprocessing and algorithm perspectives, the proposed learning machine tactfully integrates two critical techniques, that is, FN-SMOTE and cost-sensitive learning. Based on this learning machine, an online SVS assessment scheme is designed by further introducing an incremental learning strategy, which enhances the scheme's adaptability and reliability during online monitoring. Numerical test results on the Nordic test system and the real-word CSG illustrate that the proposed learning machine achieves excellent performances even if it is exposed to severe class imbalance. In addition to its reliability and adaptability during online monitoring, the learning machine exhibits desirable interpretability for SVS pattern discovery and comprehension in practical power grids. In relevant future work, an array of aspects, e.g., spatial-temporal correlation learning, advanced deep learning, and model/data mixed learning, can be explored and investigated to further enhance the data-driven SVS assessment solution's applicability in practice.