Research on the classification and evaluation of large-scale user load patterns for grid supply–demand balance regulation demands

Evaluating demand-side user loads can increase their efficiency in regulating grid supply and demand power balance. However, existing studies have not adequately considered the effects of large-scale user loads on power system balance. To address this issue, this paper proposes a method for classifying and evaluating large-scale loads on the basis of the time period regulation demand of grid power balance. First, from the perspective of the grid's time-segmented regulation demands, a “segmentation–fusion” load classification strategy that is based on time period division is designed. Second, we introduce a mean shift method that is based on the weighting of load samples to reduce the influence of the discrete load sample distribution on the clustering process. Clustering is carried out using the optimal similarity iteration method. Third, an evaluation model fitted with a back propagation (BP) neural network is proposed for evaluating massive loads. Additionally, to address the issue of load category sample imbalance, a fitted sample construction method that is based on density-aware hybrid sampling (DAHS) is introduced to improve the fitting accuracy of the BP neural network. Finally, a case study in which an actual grid and load data from a location in China are used is conducted. The simulation results validate the effectiveness and adaptability of the proposed method in evaluating massive loads. • To propose a classification and evaluation method of demand-side loads for the needs of the supply-demand balance of grids. • The "segmentataion-fusion" clustering methods can provide specific category information tailored to grid regulation periods • The mean shift clustering algorithm reduces the impact of outliers on classification results in large-scale load data. • A load-friendliness relationship fitting model based on the BP neural network is established. • The density-aware hybrid sampling method based on the load category density improves the training accuracy of the BP network.