Artificial intelligence (AI) is now part and parcel of everyday life and business. It can help make safer cars, provide cheaper healthcare, or build more intuitive appliances to use and attuned to our needs. AI might bring to mind consumer applications like self-driving vehicles, digital assistants like Siri or Alexa, or facial recognition. Still, it also has a crucial part to play in modern industry, where it can, for instance, optimise supply chains, predict maintenance needs. In our project, we are developing AI applications to support water quality management.

Water treatment plants use biochemicals to clean water. But how much chemicals should they use?   Use too little and leads to poor water quality, but too much will waste valuable resources.  The level of chemicals takes time to adjust. So, plant operators must try to predict how much water will come into the plant in the following hours to treat it effectively. Without a crystal ball, it’s hard to get that exactly right though.

Wastewater treatment plants need to plan for water from different sources. They need to process wastewater from homes and industry while also collecting stormwater in cities with combined sewage systems. The first flow is relatively predictable. People tend not to radically change their water consumption habits overnight and industry also has a certain rhythm to it. It’s much more difficult to predict rainwater though. “The flow typically follows recurring patterns that are quite unrelated to the rainfall intensities,” explains Sten Lindberg from DHI. “The rainfall-induced run-off is a complex pattern from different sources: streets, parking lots or roofs, but also parks, lawns and other green areas.”

The operator’s task of ensuring the water quality is further complicated by modern challenges.  Growing urban populations increase pressures on wastewater treatment plants. With heavy rainfall, the sewage system has to handle more water than usual and can become saturated, discharging excess untreated wastewater into nearby waterways. With the changing climate, extreme rain events are predicting to become more common, releasing more untreated water which pollutes the surrounding water bodies. Finally, the growth of urban communities adds to these challenges. Drainage must be increased when green spaces are converted to concrete.

Digital technologies like AI can support the operator in taking better decisions to adapt the level of biochemicals in the tanks. Our researchers in Copenhagen are trialling a toolbox that makes more accurate predictions on how much water a plant might need to treat, with a forecast horizon of up to 48 hours. To this end, they use machine learning to recognise patterns and relationships between data from sewers, rain gauges and weather radar information.

“This allows us to understand the relationship between rainfall and the flow of water coming into a wastewater treatment plant,” Mr Lindberg says. “We train the machine learning routines on historical time series data to predict the measures flow. Then, we use the trained model to predict the inflow rates several hours before the water flows into the facility. When we know the expected inflow in advance, treatment plants can operate more efficiently and decrease pollution”.

AI helps plants treat more water at a lower cost by helping operators to better use the existing infrastructure. As a result, this digital technology can help cities reduce the investments needed to update their water infrastructure. In addition, it makes it easier for plants to comply with EU water directives, which regulate the quality that treated water must have before to be released in the environment.