Typical Load Clustering and Feature Extraction Algorithm Considering Class Imbalance

The imbalance severely constrains the generalization ability and feature representation effectiveness of models. a research strategy for a typical load clustering and feature extraction algorithm that considers class imbalance is proposed in this paper. Firstly, a hierarchical and multi-level control architecture is constructed. By analyzing the classification characteristics of adjustable load resources, a multi-level aggregation standard technical framework is designed to achieve source-load coordinated optimization. Secondly, dynamic time warping is used to quantify the morphological differences of load curves, and a multi-agent electricity consumption behavior model is established by integrating the analysis of spatiotemporal heterogeneity under market competition mechanisms. Then, the range normalization method is employed to normalize the dimensions of typical users. Finally, simulation results show that the proposed load control strategy can ensure the tracking effect of aggregated load power to the target power, providing a reliable basis for the refined operation of power systems and demonstrating significant engineering application value.