KANMixer: A Fully Linear Network with Multi-Scale Feature Awareness for Power Quality Classification Tasks

This study introduces a novel power quality disturbance classification model, KANMixer, designed based on a fully linear KAN network with multi-scale feature perception. KAN-Mixer incorporates a multi-scale feature extraction mechanism that effectively captures the characteristics of input signals across different temporal scales. The model employs a fully linear layer architecture, founded on KANLinear, and combines piecewise polynomial interpolation with basic linear transformations to enhance the capture of nonlinear features. Experimental results demonstrate that KANMixer excels in power quality disturbance classification tasks, outperforming traditional methods across multiple evaluation metrics including accuracy, precision, recall, and Fl-score. Ablation studies further validate the effectiveness of each component of the model, particularly highlighting the significant advantages of KANLinear over traditional MLPs. This research provides an efficient and accurate new method for power quality disturbance classification, which is crucial for enhancing the safety and stability of power systems.