As the digital-water.city project enters its final months, what have our DWC partners in Copenhagen been up to?   

Our partners in Denmark have been busy testing inflow forecasts. The machine-learning (ML) and the high-fidelity model are now both up and running and can be used as decision support methods. One of the main challenges with the ML model was how to build an inflow forecast that could rely on online data when one of the data points feeding the model doesn’t work. They fixed this by building several ML models in case one or more devices didn’t send signals to the model.  

With regards to the Decision Support System (DSS), quality performance of the forecast models had to be implemented and ensured at multiple stages. For example, multiple ‘offline’ checks on the HIFI model had to be performed using historical events to make sure it could realistically simulate inflows to the wastewater treatment plant (WWTP) during dry and wet weather.  

Moreover, evaluating the operational system forecasts depends on the occurrence of wet weather events. Wastewater inflows during rain events are of main interest in the project, and thus, more complete checks on operational model performance could only be made upon availability of rainfall input and inflow measurements during rainy periods i.e. from late spring/summer in Denmark.  

Another main achievement from our partners is the web visualisation platform. The platform shows the results of models, the flow and level measurements, rain stations and rain statistics. This solution uses two-way communication: from the utility to the DWC platform, and from the DWC platform to the utility.  

With this solution, most challenges have to do with the data management. The complete collection of flow, level, and volume sensor data, data from rain gauges and weather forecast data is complex as sources, communication protocols and format vary. Since data is communicated through different sources and protocols, our partners put in place communication standards to minimize development overhead.  

What were the lessons learnt in the project? Managing a huge amount of data needs a lot of computing power and storage capacity, which can be costly in the long run. Handling big collections of data in real-time or nearly real-time puts the performance of the queries under stress when multiple users are in the system and flow predictions are run. For that reason, performance is being tested. However, prioritizing the implementation of digital solutions and set-up a business case for each of them could be a good way to fix these issues.  

Before the end of the project, our partners are looking forward to the results of the KPI’s, and seeing which model performs best in the overall running and which one is worth implementing at the WWTP. They are also looking forward to presenting the results to their stakeholders, in particular how to use the inflow forecasts in a more environmentally friendly way.