Modeling and Forecasting Short-Term Water Demand Reported by Smart Meters:

The advent of smart metering or Advanced Metering Infrastructure (AMI) for water systems generates large datasets about water demands that can provide new insight about consumer behaviors for managing water resources and infrastructure. More accurate estimates of consumption patterns and expectations of variations can be used for both planning water supply and expanding infrastructure for new development and to improve control and operation of the water systems. In this research, we explore water demand data collected at hourly intervals for a set of residential and business customers in Cary, North Carolina. Machine learning algorithms, including neural networks and regression trees, are applied to explore the influence of climate variables, including temperature, dew point and humidity, on variation in water demands. Models are evaluated based on the Mean Absolute Percentage Error (MAPE) and the standard deviation of MAPE generated over a set of trials. Times series clustering is applied to explore similar consumption patterns on datasets during weekdays and weekends. Results demonstrate the utility of applying analytic approaches and predictive models using large data sets available through smart meters.