A Novel Methodology for Microgrid Power Quality Disturbance Classification Using URPM-CWT and Multi-Channel Feature Fusion

Abstract

Addressing the limitations inherent in conventional Power Quality Disturbance (PQD) identification systems, particularly regarding the restricted information obtainable from single image features and the compromised noise immunity of single-channel networks, an innovative approach, integrating Uniform Relative Position Matrix-Continuous Wavelet Transform (URPM-CWT) and multi-channel feature fusion, is presented. This method capitalizes on the principle of feature fusion to enhance microgrid PQD identification. To begin with, each PQD signal undergoes processing through the URPM and CWT, followed by horizontal splicing to yield the URPM-CWT feature image. This is followed by the parallel deployment of three refined networks—MobileNetV2, ResNet50, and ShuffleNetV2—using the Self Fusion Module (SFM) to yield a multi-channel feature fusion classification model. The final stage involves feeding the URPM-CWT feature image into the multi-channel feature fusion classification model and applying a fully connected layer for training, leading to comprehensive perturbation recognition. Constructed using the PyTorch framework, the proposed model is evaluated on an exhaustive database of 28 distinct PQD types. In a 30db white noise environment, the method demonstrates an average classification accuracy of 99.35%, surpassing the performance of standalone deep learning recognition approaches. Simulation experiments corroborate the model’s high classification accuracy, effective recognition, and robust resistance to noise when dealing with PQD signals. Thus, the model offers promising potential for practical applications in PQD identification and classification.

Keywords

• Power quality disturbances
• deep learning
• relative position matrix
• wavelet transform
• feature fusion