Cloud-based massive electricity data mining and consumption pattern discovery

Abstract

With the development of the power systems in China, there is large volume of basic electricity consumption data accumulated. Mining these data to discover possible consumption patterns and group the users in a more finegrained way can help the State Grid Corporation to understand users’ personalized and differentiated requirements. In this work, an algorithm called TMeans is proposed to mine the electricity consumption patterns. TMeans improves the classical K-Means algorithm by presenting a set of static and dynamical rules which can dynamically adjust the clustering process according to the statistical features of the clusters, making the process more flexible and practical. Then a MapReduce-based implementation of TMeans is proposed to make itself capable of handling large volume of data efficiently. Through experiment, we first demonstrate that the consumption patterns can be effectively discovered and can be refined to very small granularity through TMeans, and then we show that the MapReduce-based implementation of TMeans can efficiently speed up the clustering process.