The classification of power quality disturbances plays a crucial role in fault warning and identification within power systems. To address the challenges posed by complex signals in new power systems, this paper proposes a neural network model that integrates deep hyperparameter convolution, multi-scale feature fusion, and ensemble learning to enhance classification accuracy. The signals are first preprocessed into two-dimensional recurrence images and input into a neural network composed of deep hyperparameter and multi-scale convolutions for feature extraction, improving feature discrimination. Finally, classification is performed using the ensemble classifier XGBoost, further boosting accuracy. Experimental results demonstrate that the model achieves high classification accuracy for various power quality disturbances, with strong noise resistance and generalization capabilities, providing new insights for intelligent grids and automated signal recognition.
