Clustering Analysis of Power Load Curves Based on KPCA and Improved K-means

The traditional power system is transforming into a new type of intelligent and digital power system. With the power load data showing high dimensionality, complexity, and diversity, the traditional load curve clustering method is facing the challenges of computational efficiency and solving accuracy. In this paper, based on the preprocessing of the original load data, the probability weights are introduced to optimize the initial clustering center of the K-means algorithm, and combined with the global convergence advantage of the improved particle swarm optimization algorithm to automatically determine the optimal number of clusters, to obtain more accurate load clustering results. On this basis, considering that the kernel PCA algorithm has the characteristic of capturing the nonlinear relationship of load data, the load data are processed with dimension reduction to reduce the computational cost. Case studies demonstrate that: Compared with the linear dimensionality reduction method and the traditional K-means clustering algorithm, the proposed method in this paper can effectively improve the overall clustering accuracy of the load curve while ensuring computational efficiency, which solves the problem of difficulty in balancing the accuracy and efficiency of the traditional clustering algorithm to a certain extent, and then provides important references for mastering the pattern of electricity consumption of power users.