Water consumption patterns as a basis for water demand modeling

Future water demand is a main consideration in water system management. Consequently, water demand models (WDMs) have evolved in past decades, identifying principal demand-generating factors and modeling their influence on water demand. Regional water systems serve consumers of various types (e.g., municipalities, farmers, industrial regions) and consumption patterns. Thus, one of the challenges in regional water demand modeling is the heterogeneity of the consumers served by the water system. When a high-resolution, regional WDM is desired, accounting for this heterogeneity becomes all the more important. This paper presents a novel approach to regional water demand modeling. The two-step approach includes aggregating the data set into groups of consumers having similar consumption characteristics, and developing a WDM for each homogeneous group. The development of WDMs is widely applied in the literature and thus, the focus of this paper is to discuss the first step of data aggregation. The research hypothesis is that water consumption records in their original or transformed form can provide a basis for aggregating the data set into groups of consumers with similar consumption characteristics. This paper presents a methodology for water consumption data clustering by comparing several data representation methods (termed Feature Vectors): monthly normalized average, monthly consumption coefficient of variation, a combination of the monthly average and monthly variation, and the autocorrelation coefficients of the consumption time series. Clustering using solely normalized monthly average provided homogeneous and distinct clusters with respect to monthly consumption, which succeed in capturing different consumer characteristics (water use, geographical location) that were not specified a-priori. Clustering using the monthly coefficient of variation provided different, yet homogeneous clusters, clustering consumers characterized by similar variation trends that were closely related to consumer water use type. The concatenation of these two Feature Vectors provided further insight into the relationship between consumption patterns and variability of consumers. An autocorrelation Feature Vector provided results that can form a basis for constructing a time-series model that is based on a group of resembling time series. The approaches presented here are steps toward utilizing the increasing amount of available water consumption data and data analysis techniques to facilitate the modeling of water demands in larger and heterogeneous regions with sufficient resolution.