A Bus Load Classification Method Based on Temporal Convolutional Attention Based Deep Embedded Clustering Model

With the rapid development of China's new power system, the proportion of distributed photovoltaic/wind power integration continues to increase, making the bus load characteristics more complex. Accurately identifying and revealing typical bus load fluctuation characteristics can provide crucial data support for power system operation and planning. Therefore, this paper proposes a bus load classification method based on the Temporal Convolutional Attention based Deep Embedded Clustering Model (TCA-DEC). First, a bus load feature classification framework based on deep embedded clustering is established, deeply mining the intrinsic characteristics of load fluctuations through unsupervised deep learning. Then, a feature encoding-reconstruction network based on temporal convolutional attention mechanism is proposed to extract significant features of different types of loads and encode them through a convolutional network. Finally, an example analysis based on actual measurement data from a city in southern China shows that the proposed method can accurately classify bus load characteristics with wind power and photovoltaic integration, improving the silhouette coefficient by 16.12%-20.51% and reducing the DB index by 23.81%-33.330/0 compared to conventional clustering methods.