DTW-based Adaptive K-means Algorithm for Electricity Consumption Pattern Recognition.

The research on electricity consumption pattern recognition generally encounters some prominent problems such as poor similarity, poor accuracy, and low efficiency of existing clustering algorithms. Therefore, this paper utilizes elbow judgment (EJ), gap statistic (GS), and DTW (dynamic time warping) to develop a DTW-based adaptive K-means (DAKM) clustering algorithm for electricity consumption pattern recognition. The algorithm includes three main aspects. First, the DTW distance with the Sakoe-Chiba band global constraint is used to find the optimal alignment between the two load curves by matching the shapes with local stretching or compression sequences. Second, gap statistic and elbow are used to obtain the optimal number of clusters for high clustering efficiency automatically. Third, a max-min DTW distance (MMDD) method is presented to optimize the initial cluster centers of the K-means algorithm. The comparative experimental results demonstrate that the proposed DAKM algorithm achieved best evaluation values of 0.7055 for DBI, 0.0237 for SSE, 132.0435 for CHI, 0.6649 for SC, and 1.1670 for DI, respectively, which proves that the proposed DAKM algorithm is far superior to other clustering algorithms. [ABSTRACT FROM AUTHOR]