A Habit Detection Algorithm (HDA) for Discovering Recurrent Patterns in Smart Meter Time Series

Conserving water is a critical problem and characterising how households in communities use water is a first step for reducing consumption. This paper introduces a method for discovering habits in smart water meter time series. Habits are household activities that recur in a predictable way, such as watering the garden at 6 am twice a week. Discovering habit patterns automatically is a challenging data mining task. Habit patterns are not only periodic, nor only seasonal, and they may not be frequent. Their recurrences are partial periodic patterns with a very large number of candidates. Further, the recurrences in real data are imperfect, making accurate matching of observations with proposed patterns difficult. The main contribution of this paper is an efficient, robust and accurate Habit Detection Algorithm (HDA) for discovering regular activities in smart meter time series with evaluation the performance of the algorithm and its ability to discover valuable insights from real-world data sets.