In recent years, data-driven methods have shown great potential for the practical application of short-term voltage stability (STVS) assessment. However, most existing research works overlook the problem of sample imbalance and overlap in STVS assessment. To tackle this issue, a novel self-adaptive data-driven method for real-time STVS is proposed in this study. First, min-redundancy and max-relevance (mRMR) is employed for feature selection to reduce the computational burden. Taking the key features as inputs, a cascaded LightGBM (CasLightGBM) model is constructed to mine STVS informatization. Based on the LightGBM and cascaded structure, CasLightGBM can enhance the assessment accuracy without sacrificing the assessment earliness. Then, focal loss (FL) is embedded into both offline and online phases of the CasLightGBM to mitigate the loss of accuracy caused by sample imbalance and overlapping, thus deriving a highly comprehensive and reliable classification model for real-time STVS assessment. Extensive numerical tests are conducted on the IEEE 118-bus system, and the simulation results demonstrate that the proposed method outperforms traditional algorithms and exhibits favorable robustness to measurement noise.
With the intensive commissioning of high voltage direct current, transient voltage problems have become increasingly prominent, which seriously threatens the safe and stable operation of the power system. On the basis of cascaded CatBoost (CasCatBoost) and sparrow search algorithm (SSA), a novel temporal‐adaptive data‐driven method for short‐term voltage stability (STVS) assessment is proposed in this paper. First, normalized mutual information feature selection is employed for important feature selection to reduce the computational burden. Taking the important features as inputs, the CasCatBoost model is constructed to mine STVS informatization. On the basis of CatBoost and cascaded structure, CasCatBoost can enhance the assessment accuracy without sacrificing the assessment earliness. Furthermore, the SSA is embedded into the offline phases of CasCatBoost to determine the optimal confidence threshold. Extensive numerical tests are conducted on the improved New England 39‐bus system, and the simulation results demonstrate that the proposed method outperforms other representative machine learning algorithms and exhibits excellent reliability, rapidity, and applicability.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.