Abstract:

The transformer vibration signal has the characteristics of being non-stationary and non-linear.This paper proposes combining the empirical mode decomposition(EMD) with adding-weight one-rank local-region method to predict the trend of the transformer vibration signal.First,the vibration signal collected at the scene is decomposed by the empirical mode.Then the phase space reconstruction theory is adopted to determine the appropriate time delay and embedding dimension for each intrinsic mode function of phase space reconstruction.After that,the adding-weight one-rank local-region method is used to predict and merge the components of the vibration signal.The experimental results show that this method can accurately predict the transformer vibration signal's change in trends and effectively monitor the transformer's running state.

Key words: transformer;empirical mode decomposition;phase space reconstruction;adding-weight one-rank local-region method;trend forecasting