Forecasting water flow in the next 24 hours
How can the Vechtstroom Water Board optimise energy use by predicting water flow?
Water boards are facing the challenge of reducing energy consumption within their infrastructure. One way to do this is to achieve an even purification process. Today, the effluence is equal to the supply flow; the same volume of water that enters the WWTP is also leaving it, which results in major issues with energy consumption.
During heavy rainfall pumps must run at full speed, which uses the highest levels of energy. However, in some cases, there are unnecessary peaks that could have been levelled off during drier periods. This can only happen if we accurately know how much water will arrive at the STP in the coming hours.
In the first phase of this project we developed a machine learning model to predict the flow of incoming water for the WWTPs in Losser and Okdenzaal up to three hours in advance. This analysed both historical data from the STPS as well as KNMI precipitation data and Vitens data on drinking water consumption. The model was then extended to increase the accuracy for up to 24 hours ahead. We later added the KNMI precipitation data and Vitens data over a longer time period, expanded the model with river and groundwater levels and also included current weather forecasts from Buienradar.
The result is a continuous, intelligent network that predicts supply flow 24 hours in advance, with 90 per cent accuracy. The model also extensively investigates the influence of seasonal trends.
The yellow area is a visualisation of the normal daily pattern of supply flow. The blue indicates the supply flow once optimised via buffering, in line with the prediction.
“The supply flow can be predicted with 90 percent accuracy, up to 24 hours ahead”
The model examines how realistic it is to buffer operational water, in collaboration with the municipalities or in installed buffer tanks. Today these are only used during peak rainfall, but could also be used in times of drought to make installations run more evenly. This would have a positive effect both on energy consumption and the maintenance needs of pumps and purification systems. Running installations at optimum frequency has the potential to postpone investment for longer.
We are applying this model in a new project for the Rijnland District Water Control Board, adapting it to create the most efficient arrangement to optimally manage its supply volumes.