A Deep Denoising and Multi-Feature Fusion Method for Power Quality Disturbance Classification

Power Quality Disturbances (PQDs) are common issues in power systems, and their accurate classification is crucial for the stable operation of power systems. However, traditional classification methods often perform poorly when dealing with PQDs signals in high-noise environments. This paper proposes a power quality disturbance classification method based on deep denoising and multi-feature fusion, leveraging the advantages of Deep Autoencoder (DAE), Fully Convolutional Network (FCN), and Long Short-Term Memory (LSTM) networks. First, the DAE module is used to denoise the noisy PQDs signals and extract effective feature information. Then, the FCN module is employed to extract the spatial features of the PQDs signals. Finally, the LSTM module, optimized with Time Series Reshape (TSR), is utilized to extract the temporal features of the PQDs signals. The spatial and temporal features are fused through a Feature Fusion Module (FM), ultimately achieving accurate classification of PQDs signals. Experimental results demonstrate that this method exhibits high classification accuracy and robustness in high-noise environments.