A power quality disturbance classification method based on multi-level optimized adaptive time-frequency analysis and deep learning

With the increasing integration of distributed generation in the power grid and the growing power quality challenges caused by various high-frequency power electronic devices, the stability of the grid is increasingly affected. In this paper, a power quality disturbance (PQD) recognition method based on a combination of time-frequency analysis and Deep Convolutional Neural Networks (DCNN) is proposed. First, the Variational Mode Decomposition (VMD) parameters are optimized using the Mean Filter Envelope Extremum Method (FE) and the Sparrow Search Algorithm (SSA). This is followed by the screening and removal of noise components from the modal components obtained through VMD decomposition. Secondly, for the selected multi-scale modal components, the Multi-Resolution S-Transform (MST) is applied to obtain the time-frequency feature maps of multiple scales, which are then input into an improved Inception-ResNet model for PQD recognition. Additionally, the Convolutional Block Attention Module (CBAM) attention mechanism is introduced to improve the model performance in capturing key information. The proposed method obtained a classification accuracy of 99.31 % in a 20–40 dB random noise environments and is validated through measured PQD data, demonstrating its reliability.