Data-driven prediction method for characteristics of voltage sag based on fuzzy time series.

• The prediction method for voltage sag events is guided by the classical fuzzy time-series prediction method. • This paper proposes a data preprocess method to eliminate redundant monitoring data and obtain better prediction performance. • This method introduces Fuzzy C-Means Algorithm to transform monitoring data into interval symbols. • This method builds a fuzzy relation based on hidden Markov to describe the relationship between disturbances and interval symbols. • This method can provide the user with the occurrence time and residual voltage information of the next voltage sag in the future. To inform the power utility and users, and help them reduce the huge financial losses due to voltage sag, it is important to obtain information on voltage sag events in advance. This paper proposes a method for predicting voltage sag characteristics based on fuzzy time series. First, we propose a homologous aggregation method to eliminate redundant data representing the same disturbance event and obtain the time series of voltage sag (TSOVS), which can describe the trend of the voltage sag data. Second, this paper introduces a fuzzification method for the time series of voltage sag based on the fuzzy c-means algorithm (FCMA), which transforms the time series of voltage sag into a fuzzy time series composed of interval symbols, to characterize the mapping relationship between the disturbance and voltage sag event. Furthermore, a hidden Markov model (HMM) of voltage sag is constructed to reveal the transformation relationship among elements in the fuzzy time series, considering the causal relationship between the disturbance and voltage sag event. Finally, the occurrence time and residual voltage of the voltage sag in the future were predicted based on this transformation relation. The measured voltage sags in a province in central China were used to verify the accuracy of the proposed method, prediction results with an accuracy of up to 90%. [ABSTRACT FROM AUTHOR]