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

In this study, a novel methodology is proposed for the intelligent detection of harmonic loads in distribution networks. The proposed method integrates parameter optimization, variational mode decomposition, and a long short-term memory network. Initially, the harmonic apparent power distortion of a nonlinear load is computed based on the IEEE Std.1459-2010 power theory. Subsequently, an arithmetic optimization algorithm is utilized to optimize the penalty parameter and the number of mode components in variational mode decomposition for the harmonic power sequence. The multimodal component sequences with robust features are subsequently chosen to transform the feature vectors of the neural network. Finally, a bidirectional long short-term memory network is utilized to dynamically capture the nonlinear features from the harmonic power sequences for the purpose of identifying harmonic loads in the distribution networks. Experimental results substantiate that this method accurately and effectively identifies harmonic power loads without the need for prior detailed information about them.