Author(s): Maria Xenochristou; Zoran Kapelan; Chris Hutton; Jan Hofman
Linked Author(s):
Keywords: Demand forecasting; Machine learning; Random Forests; Water management
Abstract: Accurate forecasts of demand are essential for water utilities in order to manage, plan, and optimize the operation of their network. This work aims to develop a new method for short-term water demand forecasting by utilizing a new data-driven approach based on Random Forests, as well as consumption recordings, household, and socio-economic characteristics, and weather data. Initial results, obtained on reallife consumption data from the UK, demonstrate the potential of this method and show the importance of disaggregating consumption when attempting to determine the influence of weather on water demand. In this study, adding weather input to the model achieved improved forecasting accuracy, especially for the aggregation of properties with medium occupancy and affluent residents during summer months.
Year: 2018