Predictive Control on the Basis of Neural Networks

Wastewater treatment plants are subject to constantly varying stress situations such as the change from day to night or the strain put on them due to rain events. Most control systems thereby react to changes in the biological parameters of the activated sludge phase such as oxygen, ammonium and nitrate.

> "Figure 1: A modern WWTP is optimized through a new Control System"

This measurement method can be problematical if, for example, heavy rain events put a strain on the system intermittently. In order to catch these surges, conventional control systems react with high concentrations of oxygen. This costs a lot of energy and can, despite this, lead to limiting values in the outflow being exceeded, if plants have been put under a lot of strain. It would be better to prepare the nitrification phase for this even before the strain occurs by raising the O₂concentration. Predictive control systems follow this approach exactly. The aim, in this case, is to save energy and to avoid ammonium peaks. A basic prerequisite for this, however, is the early capture of relevant parameters such as ammonium. For several reasons, it is not normal to measure the relevant parameters within the inflow to the wastewater treatment plant, as this would be, among other things, decidedly maintenance and cost-intensive.

For this reason, the GECO►C team uses predictive controls which operate on the basis of neural networks. Artificial neural networks are structures that are capable of learning from experience. Thus they can estimate, on the basis of the current values as well as the behaviour of the plant in the last few hours and months, which values will adjust themselves in the near future. To that end, the ammonium and nitrate values are forecast for one hour ahead and this information is processed either in an extended fuzzy controller or a feature state controller. The result is a noticeable reduction in stress peaks during the activated sludge phase, which leads to energy savings and improvements in the outflow values. Furthermore, this type of control system is also suitable for other applications at wastewater treatment plants such as for the dosing of process water.