面向含多种用户类型的负荷曲线聚类研究

Research on Load Curve Clustering With Multiple User Types

传统聚类算法在进行负荷曲线聚类时,存在不易选取初始聚类中心、需人为确定最佳聚类数、收敛速度慢等问题,并且当负荷数据中含有较多的用户类型时,其聚类效果往往较差,针对以上问题,以密度峰值聚类为基础,提出一种面向多种用户类型的负荷曲线聚类优化算法。该算法通过类间优化与类内优化的方式,实现了数据集的全局扩散与局部收敛,增强了数据的可分性,且具备一定的自愈优化能力。实验选用轮廓系数（silhouette coefficient,SC）作为聚类有效性评价指标,在国内外不同负荷数据集中进行算法的性能测试与参数摄动下的稳定性测试。结果表明该算法在面向含有多种用户类型的负荷数据集时,能够显著提高聚类有效性与鲁棒性,可为电力咨询、精准购电、负荷管理等辅助服务提供决策性信息。

Traditional clustering algorithms have many problems in load curve clustering, for example, it is not easy to select initial clustering center, it needs to determine optimal number of clusters, and convergence speed is slow. Moreover, when load data contains multiple types of users, the clustering effect tends to be poor. To solve these problems, a load curve clustering optimization algorithm for multiple user types is proposed based on density-peak clustering. The algorithm achieves global diffusion and local convergence of data sets by means of inter-and intra-class optimizations, and enhances data separability and self-optimization ability. Silhouette-coefficient is used as validity index of clustering. Performance test and stability test of the algorithm are carried out in different power datasets both from China and abroad. Results show that the proposed algorithm significantly improves clustering effectiveness and robustness for load datasets with multiple user types. In addition, the algorithm provides decision-making information for auxiliary services such as power consultation, accurate purchase of electricity and load management.