The question

How can the Stichtse Rijnlanden District Water Board stay as far below the legal discharge standards as possible and flatten out peak nitrogen concentrations?

Levels of Nitrogen and other chemicals in the effluent from sewage treatment plants (STPs) is tightly controlled as high concentrations can be damaging to the environment. The challenge is to stay as far below these levels as possible and to attempt to level off peak concentrations. To achieve this requires greater insight into the reasons behind these peaks.

The approach

Arjan Bontsema, data scientist at Ynformed, has taken up this challenge as part of his graduation thesis for Business Analytics (VU Amsterdam). Together with experts from the purification process he has assessed available data from the Zeist waste water treatment plant from both the initial process (influent, weather, season, time) and the purification process (sensor data for temperature and oxygen). He has cleaned up and translated this into usable data for a predictive model.

The solution

The result is a model that can accurately predict nitrogen concentration up to six hours in advance. This model is self-learning and, over time, it will be able to predict peaks in concentration more accurately. In addition, it incorporates interrelationships between all parameters in specific circumstances, going beyond relatively simple “if…then” assumptions, but uses facts to improve its operation.

The impact

Based on the predicted concentrations of nitrogen in the effluent and the incoming water adjustments can be made to even out concentrations. For instance, a buffer strategy can be chosen in dry weather enabling a more even inflow. These adjustments can ensure that the peak concentrations are minimized, this is not only better for the environment, flora and fauna, but also has a positive (indirect) effect on the quality of raw water. An additional by-product may be a reduction in energy consumption and maintenance needs of installations.