Classification of power quality disturbances in microgrids using a multi-level global convolutional neural network and SDTransformer approach

As the adoption of new energy sources like photovoltaic and wind power increases alongside the influx of advanced power electronic devices, there has been a significant rise in power quality disturbance events (PQDs) within power systems. These disturbances, including harmonics and voltage dips, severely impact the stability of microgrids and the efficiency of power equipment. To enhance the accuracy of identifying power quality disturbances in microgrids, this paper introduces a Multi-level Global Convolutional Neural Network combined with a Simplified double-layer Transformer model (MGCNN-SDTransformer). The model processes the input raw 1D time-series signals of power quality through multi-level convolutional and 1D-Global Attention Mechanism (1D-GAM) operations in MGCNN, which preliminarily extracts and emphasizes the key features and dynamic changes; Subsequently, the model utilizes the Multi-head Self Attention(MSA) and Multi-Layer Perceptron(MLP) components of the enhanced SDTransformer to further explore the transient local and periodic global features of the signals; The classification outcomes are then determined using a fully-connected layer and a Softmax classifier. The model effectively retains the signal's original one-dimensional temporal attributes while also delving into more complex features. This approach exhibits strong resistance to noise and enhanced generalization skills, markedly improving the detection accuracy of power quality issues within microgrids.