| 引用本文: | 刘晓峰,康进,马翔,沃建栋,吕磊炎,吴浩.基于快速动态时间弯曲和最小覆盖球的多日负荷曲线聚类方法[J].电力自动化设备,2022,42(7): |
| LIU Xiaofeng,KANG Jin,MA Xiang,WO Jiandong,Lü Leiyan,WU Hao.Clustering method for multi-day load curves based on fast dynamic time warping and minimum covering sphere[J].Electric Power Automation Equipment,2022,42(7): |
|
| 摘要: |
| 现有负荷曲线聚类的研究主要基于单日负荷曲线或多日同时刻的负荷分布开展，忽略了负荷在多日间的波动特性和负荷曲线的时间滞后特性，导致聚类结果的准确度和鲁棒性不足。综合考虑负荷的波动特性和时间滞后特性，提出一种快速动态时间弯曲和最小覆盖球相结合的多日负荷曲线聚类方法。在考虑负荷时间滞后特性的基础上，利用快速动态时间弯曲和多维尺度缩放对负荷曲线进行降维；为每个降维负荷迭代寻找最小覆盖球，并计算不同覆盖球的球间相似度；利用谱聚类算法得到相似度矩阵。算例结果表明，所提方法在准确度和鲁棒性上较传统方法有一定优势。 |
| 关键词: 多日负荷曲线 快速动态时间弯曲 最小覆盖球 谱聚类 多维尺度缩放 |
| DOI：10.16081/j.epae.202205049 |
| 分类号:TM714 |
| 基金项目:国家自然科学基金资助项目(U2066601) ；国网浙江省电力有限公司科技项目(5211JH1900M3) |
|
| Clustering method for multi-day load curves based on fast dynamic time warping and minimum covering sphere |
|
LIU Xiaofeng1, KANG Jin1, MA Xiang2, WO Jiandong2, Lü Leiyan2, WU Hao1
|
|
1.College of Electrical Engineering, Zhejiang University, Hangzhou 310027, China;2.Jinhua Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Jinhua 321016, China
|
| Abstract: |
| The current researches of load curve clustering are developed mainly based on single-day load curve or load distribution at the same time in multiple days, which ignore the inter-day load fluctuation characteristic and time lag characteristic of load curves, causing insufficient accuracy and robustness of clustering results. Comprehensively considering the fluctuation and time lag characteristics of loads, a clustering method of multi-day load curves is proposed combining fast dynamic time warping and minimum covering sphere. On the basis of considering the time lag characteristic of loads, fast dynamic time warping and multi-dimensional scaling method are used to reduce the dimension of load curves. The minimum covering sphere is iteratively found for each dimension reduced load, and the similarities between different covering spheres are calculated. The spectral clustering algorithm is used to obtain the similarity matrix. Case results show that the proposed method has certain advantages than the traditional method in terms of accuracy and robustness. |
| Key words: multi-day load curves fast dynamic time warping minimum covering sphere spectral clustering multi-dimensional scaling |
