一种基于核密度聚类的电力负荷模式精细化识别算法研究

Research on Refinement Recognition Algorithm of Power Load Model Based on Kernel Density Clustering

密度峰值聚类算法(DPC)是一种基于密度的非监督学习算法.分析用电类型复杂的电力负荷数据集时,存在负荷曲线聚类效果过分依赖人为参数设定和无法识别潜在用电模式的缺陷.结合非参数核密度估计,使用带宽搜索与边界优化提出一种适应多类型复杂用户的电力负荷数据优化聚类算法.在某市10KV真实数据集中进行算法测试,使用Davies-Bouldin有效性指标对比优化前后算法聚类效果.结果表明改进算法在面向用户类型复杂的电力数据集时,能够实现已知用电模式精确识别与潜在用电模式的深度挖掘并显著提高聚类有效性.

The density peak clustering algorithm(DPC)is a density-based unsupervised learning algorithm.When analyzing complex load datasets,the clustering effect dependent on human parameters and it is unable to identify potential electricity patterns.Using non-parametric kernel density estimation,bandwidth search and boundary optimization to optimize load clustering algorithm for multi-type complex power users.In the experiment,the algorithm was tested in a 10 KV real data set of a city,and the Davies-Bouldin index was used to compare the effectiveness of clustering before and after optimization.The results show that the improved algorithm can accurately realize the known power consumption patterns,deeply find the potential power consumption patterns and significantly improve the clustering efficiency of load curve data sets.