Data-driven Voltage Sag Source Localization Based on Kernel Naive Bayesian Classification

Aiming at the problems of low locating accuracy, data anomalies and missing in the extracted feature of the existing source localization methods, a data-driven voltage sag source localization is proposed based on multiple localization features and Kernel Naive Bayesian (KNB) classification. Firstly, in order to explore embedded information of the sag events, the causing mechanism of five type voltage sag is analyzed by simulation results, and seven kinds of localization features are extracted. These features include disturbance power and energy, impedance real polarity, system trajectory slope, current real-part polarity, and positive and negative sequence disturbance power. Then, considering that voltage sag source localization is actually a binary classification task, this paper adopts a simple and efficient kernel naive Bayesian classifier combined with multiple localization feature vectors to construct an upstream and downstream localization model of voltage sag sources for improving the accuracy and robustness of method. Finally, simulation verification is carried out in the IEEE33n ode system. The results show that, compared with common source localization methods, the proposed method has a high localization accuracy, as well as good reliably in locating various types of voltage sag sources when the feature data contains abnormal or missing data. It is applicable in complex practical applications.