Clustering and Anomaly Detection of Users Based on Seasonal Load Characteristics

Accurate implementation of load clustering analysis can improve the accuracy of forecasting future load demand, which is more helpful for power system planners to better adjust the balance of power supply and demand, and reasonably plan the construction and upgrading of power grids to meet the demand of different load types. However, current clustering algorithms have some limitations for irregularly shaped clusters and data with different density distributions. Therefore, in this study, a Density-Based Spatial Clustering of Applications with Noise for Multi-Strategy Improved Crown Porcupine Optimization Algorithm (MSCPO-DBSCAN) clustering algorithm is proposed by introducing Sobol sequence initialization population, adaptive t-distribution perturbation and adaptive chaotic sequence Gaussian mixed-variance perturbation to decompose and cluster the electricity consumption data of $\mathbf{5 0 0}$ corporate users in a year according to seasonal time scales, and the improved algorithm demonstrates superior clustering effect. Compared with the K -means algorithm and the traditional DBSCAN algorithm, the contour coefficient of the improved CPO-DBSCAN algorithm is closer to 1, which indicates that the algorithm has higher accuracy and reliability in identifying the behavioral patterns of users’ electricity consumption. Finally, the anomalous data points in electricity consumption were successfully identified by performing Local Outlier Factor (LOF) anomaly identification on the clustering results for each season.