Multimodal feature enhanced Bi-LSTM model for harmonic power load identification in distribution network

Abstract The scene of harmonic pollution in active distribution network is complex, and the data-driven harmonic load identification method needs to fully extract the nonlinear and non-stationary characteristics of harmonic power sequence signals to improve the recognition. In order to improve the intelligent identification of the harmonic loads in distribution network, a method of harmonic power load identification based on parameters optimization variational mode decomposition and long short-term memory network is proposed. Firstly, the harmonic apparent power distortion rate of nonlinear load is calculated based on the IEEE Std.1459–2010 power theory. Then, the arithmetic optimization algorithm was used to optimize the penalty parameter and the number of mode components of the variational mode decomposition of the harmonic power sequence, and the modal component sequences with obvious feature information were selected to reshape the input feature vectors of the neural network. Finally, the bidirectional long short-term memory network is used to adaptively extract the nonlinear characteristics of harmonic power sequences, and the harmonic loads in the distribution network are identified. Experimental results show that the method can accurately and effectively identify the harmonic power loads in distribution network without knowing the harmonic loads detailed information in advance.