Aiming at the problem of low accuracy in directly applying traditional clustering algorithms to time series clustering in the electricity load of universities,a DLT-Kmedoids algorithm combining DTW distance,LB_Keogh distance and time window is proposed to improve the accuracy and efficiency of clustering algorithm applied to time series.The DLT-Kmedoids algorithm uses DTW to calculate the distance between time series data instead of traditional Euclidean distance measurement,improving the accuracy of similarity measurement algorithms and also improving the accuracy and complexity of clustering algorithms.LB_Keogh distance is introduced to filter out most sequences that are unlikely to be optimal matching sequences before calculating DTW distance,and DTW is used to compare the remaining sequences one by one to further reduce the complexity of the algorithm.Finally,we analyze the time series data of electricity consumption in university buildings,and compare it with mainstream clustering algorithms.It is showed that the proposed algorithm can more accurately identify similar load patterns and perform clustering analysis with higher efficiency for the clustering task of electricity consumption data in universities.
关键词
用电负荷数据/动态时间弯曲/LB_Keogh/聚类/用电模式
Key words
power load data/dynamic time warping/LB_Keogh/clustering/power consumption mode
维普
万方数据