First of all, the estimator for the N 2-production was dependent on data from the ammonium-meter. Since this had yet to be installed there was no data to use for model calibration and validation. It could have been possible to use the N 2-production estimates calculated from the laboratory measurements, but these were so few that it would be doubtful if they would yield a good model. Secondly, the model devised had to be rather simple due to the constraints in time and resources available. Biological systems tend to be both non-linear and able to adapt to changes in the environment so any good model would necessarily be rather advanced. The models used by other authors (Michael Nielsen et al. 2005; Hao, J. J Heijnen, and van Loosdrecht 2002; Van Hulle 2005) to simulate the deammonification process was by far too advanced to be possible to adapt and implement in this project. 3.3
CHOOSING CONTROL STRATEGY
The lack of a mathematical model made selection of possible controllers rather narrow. All relatively advanced controllers based on models, such as MPC, LQ, LQG, etc., was not considered further. The theory about the deammonification process suggest that the system will be highly nonlinear and thus the PID controller would not be well suited either since it is mainly designed for LTI-systems. Instead, focus was turned to the field of extremum-seeking control. From simulations (Van Hulle 2005; Hao, J. J Heijnen, and van Loosdrecht 2002) of the process it was assumed that an optimum in N 2-production existed with respect to the DO (see Figure 6, page 12) and also that the ASL (see Figure 7, page 13) could be used to change this optimum. It was decided to start by creating a controller using only the DO-reference signal as a control-signal, while connecting the ASL in a feed-forward manner if possible. One reason for this was the available time left to complete this work and another reason was that many simpler extremumseeking control algorithms investigated was based on a single control-signal.
Figure 13 Modified system used to design the controller. Focus was first directed at different control schemes based on sinusoidal perturbations of the control-signal. These schemes typically find the optimum within a time frame similar to the system dynamics (Ariyur and KrstiÄ‡ 2003). This type of controller would most probably have been very well suited for the task but it had to be turned down for a number of reasons. The first being that the scheme was rather complex. Fully understanding and implementing the theory would exceed the available time left for this project. Another problem with complex controllers is that they are hard to explain to people without a background in control: one of the specifications the controller should meet. Also, the ammonium-meter had a time-delay of several hours which seemed to greatly exceed 21
Published on Nov 9, 2011