Extraction of Power Consumption Patterns using Non-negative Tucker Decomposition

Abstract

Electricity leveling is performed so that electricity is used more efficiently, to have a basis for medium- and long-term demand planning, and to construct stable “smart grids”; to achieve this, the amount of power used during various time slots and what it is used for must be determined in detail. Non-negative tensor factorization (NNTF) is an analysis method that extracts features from higher-order data and is useful for visualization and interpretation and has been widely used within this field. CP and Tucker decompositions have often been used with NNTF; however, Tucker decomposition seems more suitable than CP decomposition for efficiently and comprehensively understanding diversified power consumption patterns, since it is less restrictive and enables flexible decomposition. However, to the best of my knowledge, there are no prior studies where Tucker decomposition is applied to power consumption data.The objective of this study is to use Tucker decomposition to understand power consumption patterns and verify the usefulness of doing so. In this study, power consumption data is used in the form of a 4th-order tensor, and the extracted power consumption patterns are then used to perform a hierarchical cluster analysis, which is used as a basis for grouping subscribers. Finally, the results are investigated to provide a discussion of the usefulness of Tucker decomposition.