The project “Operational optimization of wastewater in Kungälv municipality” is part of Vinnova's call “Start your AI journey”, which is aimed at public organizations. The municipality of Kungälv, in consultation with Sweco and Calejo Industrial Intelligence as consultants, has looked at the possibilities of using AI in VA.
During heavy rainfall, water from precipitation, groundwater and sea penetrates the wastewater network, which can cause floods when the flow between the pumping stations becomes too high. The phenomenon is called in broad parlance for overflow.
This additional water causes congestion in the system and discharges of untreated wastewater.
Additional water is the single largest cause of flooding at sewage treatment plants, something that has negative consequences for the environment and people in terms of nutrients and the spread of infection. The authorities follow the development of nutrients very seriously and stricter emission requirements are to be expected here. This is today one of VA-Sweden's most pressing problems, says Jonas Hed, project manager at Sweco and also project manager for the project in question.
Major environmental benefits
The project has investigated whether it is possible to create a forecast for future wastewater flows with the help of an AI-based model and whether it is possible to create a more even flow with the support of such a forecast and thereby reduce the risk of overflow. Input to the model has been data from the pump stations as well as precipitation information.
The benefits of being able to produce a forecast and based on this optimize VA systems are great both financially and environmentally for both the municipality and society, says Jonas Hed.
Simulate complex systems
The model that was developed consisted of a combination of mathematical models and neural networks, so-called gray box modeling. The advantage of this modeling technique is that it is possible to improve the simulation of complex systems, where there is currently no measurement values or understanding of how they work. The model was trained and verified against historical data with good results.
Because the model can predict what happens when different parameters change, it can be used to also test a large number of different future scenarios.
- By setting up a reward system, which encourages events that are good, and punishes events that do not lead where you want, the optimization can find the best regulation of the system given each situation, says Jonas Hed.
The project has clearly demonstrated that it is possible to build an AI model, ie a digital twin, over the wastewater network and then use this model to simulate and optimize pump control in order to reduce overflow. By sending start signals from the system to force-start selected pumps, the risk of overflow can be radically reduced.
The pumping stations pump on to a treatment plant and in a next step it is advantageous to also include the treatment plant as part of the digital twin and thus reduce the risk of overflow in the entire system, Jonas Hed concludes.