Early Fault Warning Method for Distribution Networks Based on Periodic Time Series Correlation

Early fault warning in the distribution network is an effective way to mitigate the impact on power quality and safe operation of the grid. However, it is often difficult to realize in practice due to the complexity introduced by harmonics, sequence components, and other factors. Notably, early faults exhibit several distinct characteristics, such as discreteness, sparsity, and lack of periodicity in the time series, which clearly differ from normal current and voltage states, including harmonics and sequence components. Based on this inherent feature, this paper proposes a Periodic Time Series Correlation (PTSC)‐based early fault warning method for distribution networks. First, a convolutional network is incorporated into the end‐to‐end network to mitigate the influence of interference signals. Then, an extension module, a local correlation module, and a period module are proposed to focus on the global–local correlation differences of time series signals with periodic characteristics. Finally, an optimization algorithm is designed based on a set of cosine functions to model a priori periodicity information of the grid. The proposed algorithm is applied to the IEEE‐33 node grid system, and its effectiveness is verified by comparing it with state‐of‐the‐art models such as Transformer and Autoformer.