A GPU Based Parallel Clustering Method for Electric Power Big Data

• Published in: 2017 4th International Conference on Information Science and Control Engineering (ICISCE) pp. 29 - 33
• Main Authors: Cong Ji, Zheng Xiong, Chao Fang, Hui Lv, Kaizhen Zhang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.07.2017
• Subjects: Acceleration; Algorithm design and analysis; big data; Clustering algorithms; Convergence; CUDA; GPU; Graphics processing units; Industries; Instruction sets; K-means clustering; parallel computing; power load curve
• DOI: 10.1109/ICISCE.2017.16

With the explosive growth of user load data in power consumption information collection and load control systems, traditional computing frameworks and methods are faced with tremendous computational pressure when dealing with massive user load clustering and carrying out load characteristic analysis. In this paper, with a view to increasing accuracy and computational power of graphic process unit (GPU), the fast parallel K-means clustering algorithm is proposed based on Nvidia compute uniform device architecture (CUDA). This algorithm uses parallel speedup strategies, such as parallelization of computing distance between the data to be divided and the clustering center, parallelization of counting numbers of category changing curves, rational allocation of blocks, which greatly improves the clustering speed of the user load curve. A number of test examples show that K-means power load curve clustering algorithm based on CUDA proposed in this paper has a high speedup ratio and strong adaptability, which is a good way to solve the problem of massive load curve clustering.