A hidden state Markov Model for identifying the long term dynamics of indoor household water uses

Abstract

The emerging integrated water cycle management paradigm places a greater emphasis on demand-side management at the household/cluster scale than traditional design approaches. This is the motivation for the development of models that capture the dynamics of household water use at smaller spatial and temporal scales than those traditionally adopted for design of water cycle infrastructure. This study utilised data from Hunter Water Corporation (HWC) that consisted of 161 houses with measurements of monthly indoor water use over a period of 10 years. Temporal analysis of the dataset indicates that household indoor water use is mainly influenced by the household occupancy. Two major sources of occupancy dynamics were hypothesized; permanent long-term changes due to occupants permanently moving in/out and temporary short-term changes due to the occupants going on holidays and/or having visitors. In this study, a household indoor water use model was developed in which a hidden Markov model (HMM) framework was used to identity the long-term dynamics of household occupancy. The preliminary model diagnostics indicate that a reasonable fit was obtained for up to 40% of the households. Further challenges include identifying suitable drivers of the long-term dynamics and capturing the short-term dynamics evident in the data.