Classification of multiple power quality disturbances based on CNN-BiLSTM-Attention

Abstract

In order to improve the classification accuracy of power quality disturbances, a new method combining convolutional neural network (CNN) and attention mechanism bidirectional long short memory network (ABiLSTM) for power quality disturbances classification is proposed. Firstly, spatial features of disturbance signals are extracted through CNN. Secondly, the BiLSTM hidden layer learns the internal dynamic transformation rules of local features extracted by CNN and extracts temporal features; Using attention mechanism to distinguish features extracted by BiLSTM through weighting, and mining deep temporal correlations. Finally, the disturbance classification results are output through the fully connected layer. During the experiment, the power quality disturbance signal is simulated by MATLAB Simulink. The data set obtained from the simulation is used for training and testing the algorithm, and the real data set is used to verify the algorithm. The experimental results indicate that the CNN-ABiLSTM model can automatically recognize and classify features related to power quality disturbances. Compared with other methods, this method overcomes the limitations of traditional signal analysis and feature selection, and the classification accuracy of CNN-ABiLSTM for power quality disturbance signals is 99.79%, which is superior to CNN, CNN-LSTM, and ABiLSTM methods.

Keywords

• power quality
• convolutional neural network
• bidirectional long and short term memory network
• attention mechanism
• disturbance classification