Typical Power Load Profiles Shape Clustering Analysis Based on Adaptive Piecewise Aggregate Approximation

2018 3rd Asia Conference on Power and Electrical Engineering

能源学;电工学

Tang, Quan^1 ; Zhang, Quanming^1 ; Deng, Yingying^1 ; Zhu, Mi^1 ; Su, Yunche^1 ; Liu, Junyong^2 ; Wang, Xiaodi^2

Sichuan Electric Power Corporation Power Economic Research Institute, Chengdu

610000, China^1

College of Electrical Engineering and Information Technology of SCU, Chengdu

610065, China^2

关键词: Calculating time;    Cluster algorithms;    Data dimension reduction;    Piecewise aggregate approximation;    Pre-processing step;    Shape characteristics;    Shape clustering;    Similarity measurements;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/366/1/012078/pdf
DOI  :  10.1088/1757-899X/366/1/012078

来源: IOP

【 摘 要 】

With the trend of increasing dimensions of collected power load data, the data dimension reduction and classification become essential pre-processing steps for data mining and further application. On the basis of adaptive piecewise aggregate approximation(APAA) and k-Shape algorithm, a novel method is proposed. In light of the fluctuation degree and shape characteristics of the original load profiles, the new load dataset with variable temple resolution replace the old one by APAA, and further k-Shape algorithm is adopted for lower dimension load profiles clustering. K-Shape algorithm cluster curves with distance SBD as similarity measurement and also a novel method to extract the representative centroids is mentioned. The experiment testifies that APAA-kShape algorithm has shorter calculating time, higher accuracy and represent better load patterns than other cluster algorithms.