Model Based Control of Wastewater Treatment Plants

In the contribution a (non-exhaustive) overview is given on the use of mathematical models in control systems of wastewater treatment plants. As a starting point, the four building blocks of any controlled process are used, 1) the process itself, 2) the actuators that allow manipulation of the process, 3) the control algorithm that calculates a proper action for disturbance rejection or setpoint tracking and 4) the measurement devices that provide information on the important output variables and disturbances. The use of models in controller design and tuning is reviewed. Besides traditional feedback control via PID, optimal control schemes and MIMO control systems will be covered briefly. Attention is drawn to the usefulness of feedforward control strategies to compensate for the effect of predictable disturbances. The impact of the time-varying and nonlinear nature of bioprocesses on the choice and design of controllers is discussed subsequently. Adaptive and non-linear control approaches that deal with this are explained. Because measurements of disturbances and output variables form an integral part of the control loop, and a lack of adequate instrumentation is still felt as the main bottleneck for the introduction of control systems, the potential solution provided by incorporation of mathematical models in monitoring systems is presented. As a first application of models in this control system input part, the concept of software sensors will be described and illustrated with practical applications. First, the construction of state observers around process models will be introduced. Next, it will be shown how on-line estimates of model parameters can be obtained. Finally, the problem of on-line model selection will be addressed. As software sensors aim at extracting the utmost information from the scarce data, the problem of the richness of information in the data will be shortly introduced and some possible solutions will be proposed. The manner in which rich data can be obtained by excitation of the process variables using advanced control schemes or by performing so-called "In-Sensor-Experiments" will be illustrated using actual data obtained at pilotand full-scale wastewater treatment plants. A second use of mathematical models to provide data to the control algorithm of a treatment plant is the prediction of disturbances on the one hand and prediction of the allowable discharge to the receiving water. For instance, the effect of rain events, the impact of return liquors from sludge treatment or the optimal batch scheduling at production facilities, may be modelled and the model may then be used to predict future influent flows and composition over a certain time horizon. Similarly, incorporating the current and future receiving water quality dynamics in the control strategy receives increasing attention. It will be illustrated using a test case of a large city how mathematical models can be useful to predict the effluent quality required during the forthcoming period to guarantee the receiving water’s quality.