A Load Type Identification Method of Distribution Network for Under Frequency Load Shedding

Accurate and efficient load type identification determines the rationality of load cutting decision in under frequency load shedding (UFLS). However, the existing methods suffer from the shortcomings of high difficulty in dimensionality reduction training, high randomness in clustering and the clustering results are prone to getting trapped in local optima. Based on this, a load type identification method of distribution network for UFLS is proposed. In the process of dimensionality reduction, the residual module is introduced to strengthen the feature extraction ability, and the one dimensional residual convolutional autoencoder (1D-RCAE) is utilized to extract the low-dimensional features of high-dimensional load data. In the clustering process, the Canopy algorithm is used to determine the number of clusters and the initial clustering centers, combined with the spectral clustering algorithm based on the k-nearest neighbor method to achieve the division of the dataset. The example results show that the proposed method can truly and effectively reflect the actual electricity consumption characteristics of various types of loads, and provide guiding suggestions for regular adjustment of UFLS by the dispatching department.