The invention relates to a novel method for identifying
overvoltage inside a distribution network. An atomic
decomposition algorithm optimized based on an imperial colonial competition
algorithm is applied to the analysis of over-
voltage signals, and the optimized
algorithm can be used to extract the internal characteristics of signals quickly and effectively without being disturbed by
noise. A three-phase characteristic atomic spectrum is constructed based on the atomic
decomposition algorithm, waveform signals only containing time-domain information are converted into high-dimensional characteristics containing time-frequency information, and the time-frequency characteristics of
overvoltage signals are described completely. The three-phase characteristic atomic spectrum is recognized directly by using a
convolution neural network, the shortcomings of a shallow learning identification algorithm for the identification of high-dimensional characteristics are overcome, and the subjectivity and complexity in the process of characteristic reduction are avoided.
Simulation waveforms and physical models verify that the distribution network
overvoltage type identification method of the invention has high identification accuracy and strong adaptability.