Control and identification in activated sludge processes = Regeling en identifikatie in aktief-slib processen
toon extra info.
|[S.l. : s.n.]|
|1 online resource (155 p)|
|Met lit. opg. - Met samenvatting in het Nederlands toon alle annotatie(s)
|Straten, Prof. Dr. Ir. G. van ; Keesman, Dr. Ir. K.J. ; Klapwijk, Dr. Ir. A.|
|Samenvatting door auteur||
This thesis is about control and identification in activated sludge processes (ASP's). The chapters in this thesis are divided in two parts. Part I deals with the development of the best feasible, close-to-optimal adaptive receding horizon optimal controller (RHOC) for N-removal in a continuously mixed alternating activated sludge process reactor. Subsequently this controller and the most common existing controllers are mutually compared by means of simulations. In addition the application of the close-to-optimal RHOC controller to a system of two hydraulically connected alternating reactors is simulated for a range of plant designs within this class. In this way the combination of design and operation is optimized. Part II concerns identification on the basis of DO-measurements and respirometry. First the DO-dynamics in a continuously mixed ASP reactor are identified, including the non-linear relation between k L a and q air . Subsequenly the dynamics of a (DO-sensor based) continuous flow respirometer are identified by exciting its dynamics.
In chapter 1 the principles of the N-removing ASP are shortly explained. The new problem of total-N removal is discussed. The general features of the ASP control problem are listed: disturbance attenuation, storm events, process uncertainty and variation, multiple time-scales. Special attention is paid to the potential of RHOC. The literature with respect to operational aspects of N-removal as well as the use of DO-sensors and respirometers in ASP operation is coarsely reviewed. It is argued that the anoxic periods approach for N-removal offers two principle advantages over the anoxic zones approach: excitation of dynamics and no need for internal recirculation. Some problems in the field are indicated. With respect to DO-sensors it is illustrated that the challenges today are in the field of extracting not only DO but also additional information from its readings. All experiments in this thesis have been carried out at a pilot scale ASP. A description of this pilot plant is given in chapter 1. The chapter ends with the formulation of research objectives and the thesis outline.
Chapters 2 till 5 present the design procedure for the adaptive RHOC for control of NH 4 and NO x , though not exactly chronologically. The first step is presented in chapter 4, it concerns application of optimal control to the N-removal part of the generally accepted Activated Sludge Model no. 1. From this optimal control study it occurs that alternating nitrification/denitrification, as opposed to simultaneous nitrification/denitrification, may be optimal indeed. This, together with the risk of sludge bulking at limiting DO-values, justifies the limitation to alternating process operation. To implement an optimal control strategy on-line the receding horizon principle is needed, leading to RHOC. RHOC uses an internal process model for short term predictions. Hence a computationally efficient process model is required. Such a model is developed in chapter 2 by capturing the slower process dynamics in time-varying model parameters. It is taken into account that the model structure must be suited for recursive identification of the time-varying model parameters from the measurements.
RHOC, like any model predictive controller, computes the current controls on the basis of model predictions upto horizon H . Hence the sum of squared 1, 2, .., H -step ahead prediction errors is a natural identification criterion. In chapter 2 this idea is postulated and applied to NH 4 /NO x measurements collected from the pilot scale ASP described in chapter 1. H appears to affect the parameter estimates significantly, supporting the idea that use of this new identification criterion will improve MPC performance in general.
In chapter 4 RHOC with this simple model is applied to the pilot plant's alternating reactor. The controller successfully passed several tests, but it also appeared that the performance of this controller is suboptimal due to inaccurate model predictions. This was to be expected, as the simplicity of the N-removal model in chapter 2 has been achieved by capturing the slower process dynamics in the model parameters, while in this stage they are not recursively estimated.
The results of chapter 4 illustrate that recursive identification of (some of the) model parameters is required to keep the model uptodate. Chapter 3 presents the algorithm for recursive identification of those model parameters. The Kalman filter is used, because it has the attractive feature that the filter gain accompanying non-identifiable parameters ( e.g. the nitrification rate during anoxic periods) increases linearly in time. It is proven that this increase of the filter gain will not cause instability during normal process operation. The method performs excellently on real data.
Chapter 5 concerns adaptive RHOC of N-removal in alternating ASP reactors, being the combination of the recursively identified model in chapter 3 and the RHOC controller in chapter 4. Although stability of the nonlinear RHOC feedback controller has not been proven, not to mention its combination with recursive identification, only one source of instability was encountered in many experiments. This is the scenario in which NH 4 dominates the objective functional, its setpoint is zero and the estimated rate of nitrification has become negative for whatever reason. In that case the controller will keep aeration off to prevent the predicted production of NH 4 , as a consequence no new information to update the estimated nitrification rate is obtained and the deadlock is there. Obviously this scenario is easy to prevent and does not occur under normal operating conditions.
In chapter 4 the unusual observation is done that the RHOC performance is nearly invariant to its prediction horizon. This triggered a study on the cause of this phenomena and an effort to generalize the results as far as possible, the results are presented in chapter 6. It has led to the derivation of an l 1 -norm optimal state feedback controller for 2-dimensional linear time invariant systems with decoupled dynamics and a single control input.
In chapter 7 the close-to-optimal adaptive RHOC of chapter 5 and three existing control strategies (timers, NH 4 -bounds based and ORP, Oxidation Reduction Potential, based) for N-removal in continuously mixed alternating reactors are compared by means of simulation. The simulations are carried out in SIMBA TM, a commercially available application within the MATLAB/SIMULINK TMenvironment, based on the Activated Sludge Model no. 1. Drawback of simulations is that the dynamics of both the sensors and the process need to be modelled. And even the best model of the ASP is nothing but a poor resemblance of the real process. However, a fair experimental comparison of multiple controllers is impossible, not only for financial reasons. Simultaneous experimental testing would require the availability of multiple identical plants in parallel. Sequential testing on one plant would disrupt the results by changes in process conditions and influent, disabling a mutual comparison. Hence simulation is the best way to compare different control-schemes. It appears that three totally different controllers (timers, NH 4 -bounds based and adaptive RHOC) can achieve a more or less equal performance, if tuned optimally. Adaptive RHOC turns out to be superior in terms of sensitivity to suboptimal tunings. The timers approach is attractive for its simplicity, but very sensitive to suboptimal tuning.
Chapter 8 describes a simulation study with the scope to optimise the plant design and operation strategy of alternating activated sludge processes for N-removal with two hydraulically connected reactors. The methodology is to simulate the application of RHOC to a range of different plant designs within this class of systems. The RHOC algorithm is obtained by reformulating the controller of chapter 4 for a 2-reactors system. It appears that in the optimal process design the two reactors are placed in series, the first reactor is about four times as large as the second one. A conceptually simple feedback controller straightforwardly implements the improved operation strategy. The results of this chapter strongly advocate the simulation of optimal control applied to complex process models. The results are unexpected and indicate a significant outperformance of the current operation strategy. This kind of simulation studies at least serves as an ideas generator.
Chapter 9 presents a grey-box modelling approach for the identification of the nonlinear DO dynamics. Herein, singular value decomposition of the locally available Jacobian matrix, or equivalently eigenvalue decomposition of the parameter covariance matrix, as well as parameter transformation are essential techniques. The use of respiration rate measurements greatly simplifies the modelling procedure. The approach is amongst others capable of identifying the non-linear function k L a ( q air ), i.e. the relationship between k L a and the aeration input signal q air . This is especially valuable in experimental identification of the relationship between k L a ( q air ) and the design of (newly developed) aeration equipment, the use of specific carrier materials in aerated reactors, or the presence of certain detergents. After all a higher k L a at a given q air results in a higher efficiency of energy usage for aeration, and hence identification of k L a ( q air ) for newly developed equipment can yield important sales arguments.
Chapters 10 and 11 both deal with excitation of the respiration chamber dynamics in a continuous flow respirometer with the objective to extract additional information from its dissolved oxygen (DO) sensor readings. Chapter 10 is an effort to improve the accuracy of the BOD st -estimation technique developed by Spanjers et al . (1994). Contrary to expectation, the estimates still suffer from unacceptable inaccuracy due to large parameter correlation. However, a slight modification in the measurement strategy is proposed which is expected to enable more accurate estimation. The results of experiments with this modified measurement strategy are reported in chapter 11. The estimation results convincingly discourage further efforts to identify sludge kinetics and BOD st from this type of experiments.
The two other objectives of chapter 11 are the identification of the DO-sensor dynamics and the dilution rate in a continuous flow respirometer by excitation of the respiration chamber dynamics. Two separate simple procedures are presented. Both procedures consist of on-purpose in-sensor experiments succeeded by an ordinary least squares estimation step. The feasibility of both procedures is verified in experiments with activated sludge, fed with municipal wastewater. Large experimental data sets are presented, which strongly advocate the on-line incorporation of both procedures in the everyday operation of the respirometer.
In chapter 12 those conclusions drawn in the individual chapters which are of direct relevance to practitioners are summarized. Moreover some remaining ideas, which I believe are novel and likely to be succesfull, are shortly expounded in chapter 12 as well. The ideas concern: 1) Meeting N-total effluent standards by permitting elevated effluent NH 4 ; 2) Control explicitly aiming at meeting yearly averaged effluent standards; 3) The use of pH-measurements for continuous on-line tuning of timers in a timer-based operation strategy for alternating N-removal in a continuously mixed ASP reactor.
|Trefwoorden (cab)||geactiveerd slib|
WUR, Sectie Milieutechnologie