Real-time water demand forecasting system through an agent-based architecture

Water policies have evolved enormously since the Rio Earth Summit (1992). These changes have led to the strategic importance of Water Demand Management. The aim is to provide water where and when it is required using the fewest resources. A key variable in this process is the demand forecasting. It is not sufficient to have long term forecasts, but the current context requires the continuous availability of reliable hourly predictions. This paper incorporates artificial intelligence to the subject, through an agent-based system, whose basis are complex forecasting methods (Box-Jenkins, Holt-Winters, Multi-Layer Perceptron Networks and Radial Basis Function Networks). The prediction system also includes data mining, oriented to the pre and post processing of data and to the knowledge discovery, and other agents. Thereby, the system is capable of choosing at every moment the most appropriate forecast, reaching very low errors. It significantly improves the results of the different methods separately.