Real-Time Pattern-Based Water Demand Forecasting for Smart Water Networks
Short-term water demand pattern forecast is essential for cost-effective operation of distribution systems. A revolutionary short-term demand forecasting software, DemandWatch changes the way water utilities around the world leverage historical demand data to estimate future water demands for real-time, near-optimal control and management of their water distribution systems. Its primary strength lies in its adaptive demand forecasting process, which can continually update and refine estimated demand values in real time. Armed with this power, water utilities can better plan, operate and manage their water distribution systems, improve conservation measures, minimize energy consumption, meet regulatory compliance, and enhance customer service.
Water distribution network modeling is the most effective and viable way of predicting system behavior to solve a variety of design, operational and water quality problems. One critical tool for ensuring reliable network model predictions is the accurate estimation of short-term (e.g., daily) water demands. DemandWatch is designed expressly to give water utilities accurate water demand forecasts for their distribution network models. These models can help utility planners and managers understand spatial and temporal patterns of water use to optimize system operations and capital planning.
Designed for real-time applications, DemandWatch analyzes patterns from historical demand data and uses a powerful combination of Fourier transform and time-series autoregression modeling to accurately predict short-term (e.g., 24 or 48 hours) water demands. These patterns are used to identify seasonal and weekly periodicities in daily water demands as well as daily periodicities in hourly water demands. A sequential fitting of terms is automatically generated to reveal patterns between days and hourly patterns within days. Fourier transforms are used to find daily and hourly cyclical patterns. Autoregressive terms add a “short-term memory” in order to fine tune daily and hourly predictions, greatly increasing the accuracy of forecasted values. DemandWatch generates water demand forecasts at hourly intervals or at finer user-defined resolution. Innovyze IWLive Pro can readily use the output of DemandWatch to perform accurate simulations that reliably predict network operational performance over the next few hours or days. The results of these simulations can then be used for optimal operation and management of water distribution systems.
Extremely powerful and flexible, DemandWatch can also configure and calibrate network models, aided by graphs and other pertinent reporting tools. Its adaptive learning process enables it to continuously generate predictions, taking observations from the recent past and predicting demand for the near future. Even if there is a loss of observed telemetry data, DemandWatch can still generate predictions from cyclical components of the model.
Handling Abrupt Changes in Weather Condition (Temperature Changes)
DemandWatch automatically takes into account unexpected temperature changes, which is especially important during unusually very hot summer days. It models the effect of unexpected temperature changes on daily demand by taking the difference between the temperature forecast and the historical average temperature for that particular season. This difference in temperature is then multiplied by a regression factor and added to the predicted daily demand. This means that if the weather forecast is hotter than average, the predicted demand will be increased. If the weather forecast is not available, DemandWatch will use the average weather value for that day to improve the standard demand forecast. As such, DemandWatch will always ensure accurate demand forecasting using all available data.