**Nova Gorica Polytechnic, Vipavska 13, SI-5000 Nova Gorica, Slovenia. Abstract In this paper a simple control strategy is applied to and assessed on the ...
D. Vrecˇ ko*, N. Hvala* and J. Kocijan*,** *J. Stefan Institute, Jamova 39, SI-1000 Ljubljana, Slovenia **Nova Gorica Polytechnic, Vipavska 13, SI-5000 Nova Gorica, Slovenia Abstract In this paper a simple control strategy is applied to and assessed on the wastewater treatment benchmark. The controllers used in the presented control strategy are PI controllers, feedforward control and a step-feed procedure. The controlled variables are not directly the effluent concentrations but other process variables which have an effect on the effluent. The setting of set-points is also analyzed to select the values with the best performance. Set-point analysis has shown that with an optimal setting of set-points under stormy influent conditions, the achieved plant performance is also retained for rainy and dry influent conditions. The evaluation of plant performance indicates that with the proposed control strategy, a lower number of effluent violations has been achieved, as well as lower energy consumption and lower sludge production, when compared to results published up to now. Only the effluent quality criterion deteriorated. Keywords Activated sludge process; benchmark; control; set-point analysis; wastewater treatment plant
Introduction
Water pollution is becoming one of the world’s most serious environmental problems. To deal with this problem, a lot of many wastewater treatment plants with complex configurations and technologies have been invented and put into operation. However, the increase in plant complexity also requires a higher level of process control. Various control strategies for wastewater treatment plants have been suggested (e.g. Briggs, 1967; Busby and Andrews, 1975; Londong, 1992). However, control strategies have often been tested for different plant configurations and under different operating conditions. They are therefore difficult to compare. In order to compare different control strategies fairly, the Working Group No. 1 within COST (Cooperation in Science and Technology) programme 624 has developed a wastewater treatment benchmark. The benchmark is a platformindependent simulation environment defining a wastewater treatment plant layout, a simulation model, influent data, test procedures and evaluating criteria for the comparison of control performance (Alex et al., 1999, 2000; Pons et al., 2000). An already performed analysis of the benchmark (Vrecˇ ko et al., 2001) has shown that the benchmark is a non-linear and multivariable process with strong interactions between process variables. The benchmark therefore seems to require complex control. However, the control problem can be simplified since some sub-processes in the benchmark have main time constants of different orders of magnitude (Olsson and Jeppsson, 1994). These sub-processes are weakly coupled and can therefore be controlled independently from each other if the controllers are tuned in such way as to preserve the time constants of different orders in the closed-loops as well (O’Reilly and Leithead, 1991). The control problem can be additionally simplified if sub-processes are chosen which can be successfully controlled with linear controllers. In this particular case this can be achieved if the chosen controlled variables are not directly the effluent concentrations but other process variables which have an effect on the effluent. Such a simple control strategy is applied on the benchmark in this paper. The selected sub-processes of the plant are controlled independently with PI controllers. To improve
Water Science and Technology Vol 45 No 4–5 pp 127–134 © IWA Publishing 2002
Wastewater treatment benchmark: what can be achieved with simple control?
127
D. Vrecˇko et al.
plant performance in relation to effluent standards, feedforward control and step-feed procedure are also applied. As the controlled variables in the proposed control scheme are not the effluent concentrations, special care is devoted to the optimal setting of those set-points which give the best plant performance. The paper is organised as follows. In the next section the benchmark is presented, followed by an applied simple control strategy of the benchmark. An analysis of set-points is then described. After that the presented control strategy is assessed and compared with two other control strategies. Some conclusions are drawn at the end. The benchmark
The benchmark plant layout is shown in Figure 1. The plant is designed as an activated sludge process removing organic and nitrogen compounds from wastewater. The plant consists of the anoxic zone (first two basins), the aerobic zone (last three basins) and a settler. In the benchmark plant layout, Q0 and Qe represent influent flow and effluent flow respectively. To represent the biological processes in the benchmark, the ASM1 (Henze et al., 1986) model was used, while for the settling processes the Takács ten-layer model (Takács et al., 1991) was selected. In the benchmark there are four main manipulated variables: KLa in basin five (KLa5), internal recycle flow (Qa), external recycle flow (Qr) and waste sludge flow (Qw). For the observed output variables the following effluent concentrations were selected: total nitrogen (Ntot,e), CODe, ammonia (SNH,e), suspended solids (SSe) and BOD5,e. Influent data for the benchmark is available in three influent files containing 14day dynamic influent data for different weather conditions, i.e. the dry weather file, the rainy weather file and the stormy weather file. The model of the benchmark is available in different simulation platforms such as Matlab–Simulink, GPS-X, Fortran, etc. (Alex et al., 2000). In our case the Matlab–Simulink simulation platform was used. To calculate the benchmark performance, the benchmark first has to be run to the steady state by simulating the plant at the defined constant influent. Then the simulation continues by twice applying one of the dynamic weather influent files. The performance of the benchmark is evaluated for the last seven days of simulation and includes different criteria such as effluent quality, aeration energy, pumping energy and sludge production. Control algorithms applied to the benchmark
The benchmark process is 1a non-linear and multivariable process with strong interactions between process variables (Vrecˇ ko et al., 2001). It therefore requires a certain effort to achieve successful control design. The control problem of the benchmark can be simplified if the controlled variables are not the effluent concentrations but other process variables. The idea is to consider such sub-processes, which have main time constants of different orders of magnitude (from minutes to days). Such sub-processes are weakly coupled and can be controlled independently if the controllers are tuned in such way to preserve the time Qe
Q0
Qa Qr 128
Figure 1 Benchmark plant layout
Qw
Step-feed
D. Vrecˇko et al.
constants of different orders of magnitude in the closed loops as well (O’Reilly and Leithead, 1991). The control problem can be additionally simplified if controlled variables are used which can be easily measured and successfully controlled with linear controllers. The most appropriate linear controllers for controlling the plant are PI controllers because ´ they are very simple and most often used in practice (Åstrom and Hägglund, 1995). One of the variables in the benchmark, which can be controlled with a linear PI controller, is the dissolved oxygen concentration in the fifth basin (SO,5). This variable can be controlled by using KLa5 as a manipulated variable (Briggs, 1967). The main time constant for this sub-process is a few minutes (Olsson and Jeppsson, 1994). The purpose of this control loop is to supply so much oxygen that maximum nitrification and organic matter biodegradation is achieved. The next controlled variable is the suspended solids concentration in the fifth basin (SS5), which can be controlled with a PI controller and waste sludge flow (Qw) as a manipulated variable (Lech et al., 1978). The main time constant for this sub-process is several days (Olsson and Jeppsson, 1994). This control loop is needed for retaining the desired amount of biomass in the basins of the plant. The third control loop is also based on a PI controller, which controls the nitrate concentration in the second basin (SNO,2) with internal recycle (Qa) as a manipulated variable (Londong, 1992; Galarza et al., 2000). The time constant of this sub-process is around an hour (Olsson and Jeppsson, 1994). The aim of this control loop is to supply so much nitrate from the aerobic zone to the anoxic zone that maximum denitrification is achieved. The external recycle flow (Qr) is also one of the manipulated variables. Its value is determined by simple feedforward control, which sets the values of Qr to rQ0 (Andrews, 1974), where r is a chosen constant. The aim of this control is to additionally adjust the amount of biomass in the basins depending on the influent flow. Finally, a step-feed procedure was also used. This procedure is active only when the influent flow increases above the value chosen as twice the average flow calculated from the dry influent conditions. When the above condition is fulfilled, the influent flow is redirected from the first to the second basin. The aim of this procedure is to avoid settler overload in the case of high hydraulic load (Busby and Andrews, 1975; Automated Process Control Strategies, 1997). For a better presentation, Figure 2 shows the benchmark layout with all selected control loops and control strategies. In PI controller loops, anti-windup protection was also used in order to avoid oscillations resulting from limited magnitude of manipulated variables (Alex et al., 1999). To achieve the desired performance, the controller parameters have to be properly tuned. The parameters of the oxygen and nitrate PI controller were set as defined in the basic control strategy for the benchmark. The parameters of the suspended solids PI controller were tuned by using IMC tuning rules as recommended by Olsson and Newell
Qe
Q0
SNO,2
Qa SS5
SO,5 SNO,2
set
PI
Qw
Qr
Q0 r
SO,5
set
PI
SS5set
PI
Air
Figure 2 Benchmark layout with selected control loops and control strategies
129
D. Vrecˇko et al.
(Olsson and Newell, 1999). The parameters (proportional gain Kp and integral time constrant Ti) of all PI controllers are shown in Table 1. By using PI controllers with the parameters shown in Table 1, the obtained closed-loop time constants are approximately the same as the integral time constants of PI controllers. The latter are approximately the same as the main time constants of the sub-processes. The sub-process time constants are therefore preserved in closed-loops as well, which is the condition for the sub-processes to be controlled independently (O’Reilly and Leithead, 1991). When the nitrate was controlled, the nitrate sensor was used in the control loop. The nitrate sensor characteristic is defined by the benchmark. Since the suspended solids control loop is very slow, and since it was desirable to make this control insensible to hourly fluctuations, the suspended solids concentrations used in the control loop were averaged daily. Setting the optimal set-points
As already mentioned, the controlled variables in the proposed control scheme are not the effluent concentrations but other process variables, which however have an effect on effluent concentrations. It is therefore especially important to properly set the set-points in order to gain maximum efficiency of the presented control scheme. To obtain the optimal values of set-points, an analysis was performed where all possible set-point combinations in the sets below were tested: SO, 5 set (g / m 3 ) Π{0.5, 1, 1.5, 2} SS5 set (g / m 3 ) Π{4000, 4250, 4500, 4750, 5000} SNO, 2 set (g / m 3 ) Π{0.5, 1, 1.5, 2} r Π{0.5, 0.75, 1, 1.25, 1.5}
(1)
where SO,5set, SNO,2 set, SS5 set represent the values of oxygen, nitrate and suspended solids set-points respectively, while r is the ratio between the external recycle and influent flow. The sets include possible set-point values for each controller. The ranges of the set-point values were selected in accordance with the values usually used in practice, while the number of set-points in each set was chosen so that the number of all possible combinations did not increase too much. With the sets of set-points defined in (1), a total of 400 combinations was achieved. Each combination was tested by simulation in accordance with the described procedure under different influent conditions. At a given combination of set-point values, the process behaviour was evaluated as satisfactory if all effluent concentrations were below the defined limits, except effluent ammonia (under all influent conditions) and suspended solids (under stormy influent conditions), which have to meet the following conditions: dry influent : %Tviol SNH , e < 6 % rainy influent : %Tviol SNH , e < 6 % (2)
stormy influent : %Tviol SNH , e < 6% and %Tviol SSe < 3.5% Table 1 Parameters of PI controllers Parameter
130
Kp Ti
Oxygen controller
Nitrate controller
Suspended solids controller
500 d–1.(g.m–3)–1 0.001 d
15000 (m3d–1).(g.m–3)–1 0.05 d
–0.2 d–1.(g.m3)–1 18 d
SO,5set = 2, SNO,2set = 2, SS5set = 4500, r = 1.5.
D. Vrecˇko et al.
where %Tviol SNH,e is the percentage of time the effluent ammonia limitation was violated, and %Tviol SSe is the percentage of time the effluent suspended solids limitation was violated. The above constraints were chosen by simulation so that under each influent condition, at least some set-point combinations fulfilled the above criteria. The acceptable set-points which fulfil the described effluent constraints under different influent conditions are shown in Figure 3. Each diagram in Figure 3 represents only two set-points in a set-point combination. The regions represent the acceptable set-point values according to criteria (2). One has to mention here that the simulations were performed only at the setpoint values defined in (1) (they are also specially indicated on the axes). The shapes of acceptable set-points regions are thus only approximate. More accurate shapes could be obtained by increasing the number of values in the sets of set-points. Figure 3 shows that the acceptable set-points under stormy influent conditions are a subset of the acceptable set-points under rainy influent conditions, which are again a subset of the acceptable set-points under dry influent conditions. This means some set-point combinations exist, which fulfil the defined effluent constraints under all influent conditions. Among these combinations, the following set-points were chosen as the optimal choice in respect of the effluent violations criteria: (3)
These set-point values were also chosen to evaluate the benchmark performance when controlled with the presented control strategy. The performance of the presented control strategy
The proposed control strategy with optimally selected set-points (3) was tested under dry, rainy and stormy influent conditions. Simulations were performed as specified in the benchmark. The obtained responses of the selected effluent concentrations are shown in Figures 4–6. These figures show the last seven days of simulation and were obtained under different influent conditions. From the figures it can be seen that with the proposed control scheme, the effluent concentrations were almost all below the defined limits (in the figures, the limits are the highest values on the y-axes, except for effluent ammonia, where the limit is specially indicated). The only exceptions are ammonia and suspended solids 1.5
4750
1.25
)
5000
4500
r
set
3
1.5
1
4250
0.5 0.5
1
SS5
SNO,2set (g/m3)
2
1
2
4000 0.5
0.75
.5
0.5 0.5
2
1.5
4750
1.25
1.25
4500
1
r
0.75
4250
4000 0.5
r
1.5
SS5set (g/m3)
5000
1
1.5
SNO,2set (g/m3)
2
0.5 0.5
Dry influent
1.5
3
2
1
0.75
.5
) ent
0.5 4000
2
4250
4500
4750
SS5set (g/m3)
5000
Stormy influent
Figure 3 Acceptable set-point regions under different influent conditions
131
D. Vrecˇko et al.
concentrations. Ammonia violated the defined limit under all influent conditions, while the suspended solids concentration was above the limit only under stormy influent conditions. The plant performance was also numerically evaluated under all influent conditions in relation to the criteria defined by the benchmark. These criteria are effluent quality (EQ), average daily sludge production for disposal (Pdisp_sludge), aeration energy (AE), pumping energy (PE), number of times the limit was violated (Nb Viol.) and percentage of time the limit was violated (% Tviol). The values of the criteria in the case of controlling the plant with the presented control strategy are shown in Table 2. They are also compared with two other control strategies which have been applied so far to the benchmark and reported in the literature (Singman, 1999; Rehnström, 2000). In the latter two cases more complex algorithms were used, such as external carbon control, cascade oxygen and ammonia control, and combined non-linear feedforward and feedback control of nitrate. From the results in Table 2 it can be concluded that with the simple control strategy presented a lower number of violations were obtained than in the other two cases. Moreover, the energy consumption for aeration and pumping was reduced, as was sludge production. In the case presented only the effluent quality criterion is lower, which is due to the fact that 18
100
5
17.5
4.5 90
16.5 16 15.5 15 14.5 14
4 3.5
SNH,e (g/m3)
CODe (g/m3)
Ntot,e (g/m3)
17
80
70
60
3 2.5 2 1.5 1
50
13.5 13
0.5 7
8
9
10
11
12
13
40
14
7
8
9
time (d)
10
30
10
28
9
BOD5,e (g/m3)
26
SSe (g/m3)
11
24 22 20 18
13
0
14
7
8
9
10
11
12
13
14
time (d)
8 7 6 5 4
16
3
14 12
12
time (d)
7
8
9
10
11
12
13
2
14
7
8
9
time (d)
10
11
12
13
14
time (d)
Figure 4 Effluent variables for the last seven days of simulation under dry influent conditions
100
18
12
10
8
5
80
3
SNH,e (g/m )
CODe (g/m3)
Ntot,e (g/m3)
14
6
6
90
16
70 60 50
8
9
10
11
12
13
14
30
7
8
9
time (d)
10
10
28
9
3
BOD5,e (g/m )
SSe (g/m3)
2
12
13
14
0
7
8
9
time (d)
26 24 22 20 18
10
11
time (d)
8 7 6 5 4
16
3
14
7
8
9
10
11
time (d) 132
11
30
12
3
1
40
7
4
12
13
14
2
7
8
9
10
11
12
13
14
time (d)
Figure 5 Effluent variables for the last seven days of simulation under rainy influent conditions
12
13
14
100
18
12
10
80
SNH,e (g/m3)
CODe (g/m3)
Ntot,e (g/m3)
4
14
70 60 50
3.5 3 2.5 2 1.5 1
8
6
5 4.5
90
16
40
7
8
9
10
11
12
13
30
14
0.5 7
8
9
10
11
40
13
14
0
7
8
9
10
11
12
13
14
time (d)
10 9
3
BOD5,e (g/m )
35
SSe (g/m3)
12
time (d)
D. Vrecˇko et al.
time (d)
30
25
20
8 7 6 5 4
15
10
3
7
8
9
10
11
12
13
14
2
7
time (d)
8
9
10
11
12
13
14
time (d)
Figure 6 Effluent variables for the last seven days of simulation under stormy influent conditions Table 2 Plant performance obtained using different control strategies under different influent conditions Performance criterion
EQ (kg.d–1) Pdisp_sludge (kg.d–1) AE (kWh.d–1) PE (kWh.d–1) Nb Viol. Ntot,e Nb Viol. SNH,e Nb Viol. SSe % Tviol Ntot,e % Tviol SNH,e % Tviol SSe
Dry influent conditions
6,982 2,158 7,232 2,017 0 4 0 0 3.7 0
5,8221 2,526 8,040 2,943 0 4 0 0 9.8 0
5,2712 2,661 8,127 3,973 0 5 0 0 13.4 0
Rainy influent conditions
8,515 1,978 7,274 2,560 0 4 0 0 4.3 0
7,7331 7,3672 2,393 2,494 8,058 8,104 3,007 3,672 1 0 6 6 0 0 0.7 0 15.9 15.9 0 0
Stormy influent conditions
7,713 2,288 7,303 2,366 0 4 2 0 4.5 3.3
6,7361 2,688 8,111 3,018 0 6 2 0 18 2.5
6,2462 2,825 8,149 3,827 0 6 2 0 19.9 1.3
1 Singman (1999) 2 Rehnström (2000)
the effluent concentrations were in this case closer to the limit values than in the other two cases. Conclusions
In this paper a simple control strategy is presented for controlling the wastewater treatment benchmark. The strategy for controlling this – otherwise multivariable and nonlinear – process is based on the main time constants of the different orders of magnitude of the plant sub-processes. This enables control of each sub-process independently from the other subprocesses, if the controllers are tuned in such a way as to preserve different time constants in the closed loops as well. The controllers used in the presented scheme are PI controllers, feedforward control and a step-feed procedure. The controlled variables are not directly the effluent concentrations but other process variables which have an effect on output performance. In such a control scheme the choice of set-points is especially important. Analysis has shown that with the optimal setting of set-points under stormy influent conditions, the achieved plant performance is also retained under rainy and dry influent conditions. An evaluation of plant performance at optimal set-point values has shown that, with the proposed control scheme, a lower number of effluent violations is achieved, as well as lower energy consumption and lower sludge production, when compared to results published up to now. Only the effluent quality criterion deteriorated. Our work will continue with possible improvements to the controllers and with an analysis of their influence on the overall plant performance.
133
Acknowledgements
We would like to thank Dr Ulf Jeppsson, IEA, Lund Institute of Technology, for letting us use his Simulink benchmark simulation platform. References D. Vrecˇko et al. 134
Alex, J., Beteau, J.F., Copp, J.B., Hellinga, C., Jeppsson, U., Marsili-Libelli, S., Pons, M.N., Spanjers, H. and Vanhooren, H. (1999). Benchmark for Evaluation Control Strategies in Wastewater Treatment Plants, ECC’99, Karlsruhe. Alex, J., Beteau, J.F., Copp, J.B., Dudley, J., Dupont, R., Gillot, S., Jeppsson, U., LeLann, J.M., Pons, M.N. and Vanrolleghem, P.A. (2000). The COST Simulation Benchmark: Description and Simulator Manual. Andrews, J.F. (1974). Dynamic Models and Control Strategies for Wastewater Treatment Processes, Water Research, 8, 261–289. Åstrom, K. and Hägglund, T. (1995). PID Controllers: Theory, Design and Tuning, Instrument Society of America. Automated Process Control Strategies (1997). Special publication. Water Environment Federation. Briggs, R. (1967). Monitoring and Automatic Control of Dissolved Oxygen Levels in Activated Sludge, Effluent and Water Treatment Convention, London. Busby, J.B. and Andrews, J.F. (1975). Dynamic Modelling and Control Strategies for the Activated Sludge Process, Journal WPCF, 47(5). Galarza, A., Ayesa, E., Linaza, M.T., Rivas, A. and Salterain, A. (2000). Application of Mathematical Tools to Improve the Design and Operation of Activated Sludge Plants. Case Study: The New WWTP of Galindo-Bilbao, Watermatex 2000, Ghent, September 18–20. Henze, M., Grady, C.P.L., Gujer, W., Marais, G.v.R. and Matsuo, T. (1986). Activated Sludge model No.1, IAWQ Scientific and Technical Report No.1, IAWQ, London. Lech, R.F., Grady, C.P.L., Lim, H.C. and Koppel, L.B. (1978). Automatic Control of the Activated Sludge Process–II. Efficacy of Control Strategies, Water Research, 12, 91–99. Londong, J. (1992). Strategies for Optimised Nitrate Reduction With Primary Denitrification, Wat. Sci. Tech., 26(5), 1087–1096. Olsson, G. and Jeppsson, U. (1994). Establishing Cause-Effect Relationships in Activated Sludge Plants – What Can be Controlled? In: Workshop Modelling, Monitoring and Control of Wastewater Treatment Plants, Med. Fac. Landbouww., Univ. Gent, 2057–2070. Olsson, G. and Newell, B. (1999). Wastewater Treatment Systems, Modelling, Diagnosis and Control, IWA Publishing, London. O’Reilly, J. and Leithead, W.E. (1991). Multivariable control by “individual control design”, Int. J. Control, 54(1), 1–46. Pons, M.N., Spanjers, H. and Jeppson, U. (2000). Towards a Benchmark for Evaluating Control Strategies in Wastewater Treatment Plants by Simulation, Escape 9, Budapest. Rehnström, A. (2000). Automatic Control of an Activated Sludge Process in a Wastewater Treatment Plant – A Benchmark Study, MSc. Thesis, Uppsala. Singman, J. (1999). Efficient Control of Wastewater Treatment Plants – A Benchmark Study, MSc. Thesis, Uppsala. Takács, I., Patry, G.G. and Nolasco, D. (1991). A dynamic model of the clarification thickening process, Water Research, 25(10), 1263–1271. Vrecˇ ko, D., Hvala, N., Kocijan, J. and Zec, M. (2001). System Analysis for Optimal Control of a Wastewater Treatment Benchmark. Wat. Sci. Tech. 43(7) 199–206.
