Do time series models contribute towater demand clustering?

The grouping of water demands for water distribution system (WDS) simulation is a common practice widely adopted for a variety of applications including model calibration, water quality simulation and optimal control. While the importance of water demand parameterization is undeniable, demand groups are usually defined based on the typical consumer types: residential, commercial or industrial. However, little attention has been given to the development of methodologies that generate spatial clustering solutions for demand estimation. The current study contributes to future clustering procedures by empirically testing the potential of time series models to assist with cluster identification. To achieve that goal, a comprehensive test case was generated including an increasing number of additional flow meter locations and a set of cluster cases. The results suggest that the time series likelihood alone is not capable of identifying a known clustering scenario. However, the use of the time series likelihood, as one type of criterion in cluster development, is still suggested to assure forecast accuracy.