| Abstract |
With the quick development of smart cities, data mining, feature extraction, identification and classification of different energy user groups are widely used to adjust the construction of distribution networks and power supply service strategies. This paper studies the principles and steps of the spectral clustering algorithm, considering the similarity measurement of load based on the load numerical characteristics, load curves volatility and trend characteristics. Then, the spectral clustering based on the proposed similarity measurement is performed on the substation-level power load data, and some typical load curves are categorized and analyzed. The results show that the composition of typical load curves of different substations can describe the composition of energy users in each substation. Besides, the spectral clustering method shows better performance in calculation speed, effectiveness and stability compared with k-means clustering, © 2021 IEEE. |
