Nutrition Community Era College Wastewater

60 J. Rojas, T.
Zhelev / Computers and Chemical Engineering 38 (2012) 52–63
(a) x2 x8 x4 x5 x7 (b)
Dissolved oxygen (mg/l)

2 0.5
Concentrations
0.4
1.5
0.3
1
0.2
0.5
0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Time (days) Time (days)
(c) 72 (d) 12 400
Aeration flowrate (vvh)

350
Temperature (ºC)
10
OUR (mgO2/l h)
70 300
8 250
68
200
6
66 150
4 100
64 2 50
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Fig. 5. Optimal solution for CS1: (a) non-dimensional concentrations, (b) dissolved oxygen concentration, (c) temperature, and (d) aeration flowrate and OUR (N = 20). The
optimal energy requirement is 0.56 kWh/kg (22% improvement).
Let us consider a series of nb consecutive batches for a given Fig. 5 displays the optimal solution found for CS1 with an energy
control vector u, a given initial condition of the first batch x10 , and requirement of 0.56 kWh/kg which represents a 22% improvement.
fixed load and characteristics of the influent sludge (i.e., no dis- As it can be seen from the non-dimensional concentrations in Fig. 5a
turbances). Generally speaking, in this scenario the state variables (see initial condition and input in Appendix A), hydrolysis and
will tend to be non-periodic for the first batches in the series and stabilisation are the rate-limiting factors of the reaction, that is,
quasi-periodic for the later batches (at least for sufficiently large rVS (tf ) = 0.38. This was the case for almost all optimal solutions.
nb -values). Consequently, it can happen that some of the path con- This tendency is consistent with our previous finding that the sta-
straints are satisfied for the first batches in the series but not for bilisation process is generally the limiting factor of the reaction
the last ones, or vice versa. Thus, if we evaluate the path constraints (Rojas et al., 2010). Fig. 5b and c shows the optimal DO and tem-
based on one single batch, chances are high that we obtain some perature profiles. DO is consistently below 1 mg/l despite the high
solutions that are not feasible. To surmount this difficulty, for a aeration levels, meaning that the OURs are very high. Operating
given iteration within the optimisation loop, the code evaluates temperatures are high enough to exceed pasteurisation require-
the path constraints based on the last of nb consecutive batches ments (LP (tf ) 1). This was found to be the case in most optimal
using for all of them the same control vector u. This procedure will solutions. Fig. 5d displays the optimal profiles of the aeration
ensure the periodicity of the solutions and prevent us from obtain- flowrate (N = 20) and OUR. It can be seen that the values of the
ing unfeasible solutions. According to our experience, it is sufficient aeration are generally much higher than those used in the plant
to set nb equal to or higher than 10 to ensure periodicity. Thus, in (up to 12 vvh). Even though there is an apparent randomness in
the following implementations the value 10 was chosen. its behaviour, the general tendency of the aeration flowrate is to
It should also be noted that for simplicity the reactor loading was start from higher values and to gradually decrease until reach-
regarded as instantaneous. This implies a constant volume of sludge, ing the lower boundary at the end of the reaction. It is unclear,
and allows us to implement the optimisation in a computationally though, what may have caused the algorithm to choose this rela-
much more efficient manner. tively erratic and “spiky” path. However, the qualitative behaviour
of the aeration flowrate is biologically plausible: at the beginning
7.2.1. Optimisation results CS1 of the reaction (i.e. right after reactor loading) the concentration
The final time (days) was bound within the interval [0.7, 1.5], of readily biodegradable organic matter x2 is highest; this drives
and the aeration flowrate (vvh) within [0.8, 15]. The reason for the the system to a high demand of oxygen needed to oxidise the sub-
lower boundary of the final time is that the actual plant has a max- strate; in order to match the high oxygen demand, the aeration
imum sludge production of ∼42 m3 /day; with a 10% displacement flowrate has to be higher at the beginning of the batch. As time
of the reactor content after each batch (i.e., 30 m3 ), this would allow passes and substrate is gradually consumed, the demand for oxy-
a minimum reaction time of 0.7 days. The boundaries of the aer- gen decreases and so does the aeration flowrate. The behaviour
ation flowrate have been chosen so as to allow wide variations, of the OUR as well as that of DO evidently represent a reaction
much higher than those seen in the actual plant (i.e., [0.8, 1.4]). to changes in the aeration flowrate. Thus, the qualitative optimal
The lower boundary was kept at 0.8 vvh, as a minimum, nonzero profile of the aeration flowrate makes sense of the underlying bio-
level of aeration is required in this type of equipment. Note that logical system. This result also shows that the optimal behaviour of
the power-aeration relation is given by P = 28 + 6 · qa , where P is the the aeration flowrate is far from invariable, as it is typically found in
power (kW) at the aeration level qa (vvh). This relation holds in the conventional ATAD systems. At this stage, it can thus be seen that
aeration interval [0.8, 1.4] and it was assumed that it also holds in exploiting the aeration flowrate as an additional degree of free-
the interval [0.8, 15]. dom for optimisation is a crucial way of obtaining better solutions,
J. Rojas, T. Zhelev / Computers and Chemical Engineering 38 (2012) 52–63 61
(b) 1
x2 x8 x4 x5 x7
R1
(a) R2

0.8
1.5
Concentrations
0.6
1
0.4
0.5
0.2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

(c) 60 (d) 10 500

55 8 400
Temperature (ºC)
R1
h)
R2
OUR (mgO/l
2
50 6 300
45 4 200
40 2 100
35
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Fig. 6. Optimal solution for CS2: (a) non-dimensional concentrations (solid lines: Reactor 1; dashed lines: Reactor 2), (b) dissolved oxygen concentration, (c) temperature,
and (d) aeration flowrate and OUR (N1 = 10, N2 = 10) (solid lines: Reactor 1; dashed lines: Reactor 2). The optimal energy requirement is 0.99 kWh/kg (18% improvement).
and that invariable aeration is not generally suitable for ATAD sys- as optimisation variable within the interval [0, 30]. Note that this
tems. polymer is added to the influent sludge which has an original con-
The optimal reaction time tf was found at the lower boundary centration of 15 g/l. Thus, after pre-thickening the minimum and
of 0.7 days (the typical time of operation being 1 day). It was found maximum VS concentrations are 15 and 45 g/l, respectively.
that the algorithm tends to favour shorter final times, which is also The optimal solution found for CS2 is displayed in Fig. 6 with an
consistent with the aforementioned sensitivity analysis. Thus, the energy requirement of 0.99 kWh/kg which represents an improve-
optimal value of the final time, being the value of the lower bound- ment of 18%. The concentrations in Fig. 6a (see initial condition and
ary, makes sense of the dependence of the energy requirement input in Appendix A) show that hydrolysis and stabilisation are,
(objective functional) on the final time: shorter final times tend to once more, limiting. As in the previous case study, DO in Fig. 6b is in
result in lower energy requirements. Let us recall that conventional both reactors consistently below 1 mg/l. Even though the temper-
systems use a 1-day final time due to operational convenience, thus ature of the first-stage reactor is relatively low, the second-stage
not exploiting this important degree of freedom. Even though in reactor operates consistently at thermophilic temperatures (see
some optimal solutions tf was between the boundaries, in most (and Fig. 5c), and pasteurisation requirements are exceeded. The overall
actually the best) of them it was found at the lower boundary. This low temperatures are a result of the low optimal polymer concen-
seems to suggest that, when minimising the energy requirements, tration (6.5 g/l).
lower reaction times and higher aeration flowrates are favoured. The optimal aeration profiles (N1 = 10, N2 = 10) in Fig. 6d show
It should also be noted that the objective function is multimodal that, with exception of “spiky” intervals, the aeration levels are
as several, diverse solutions were found. Thus, by no means we can generally lower than in the current operation of the plant. It does
guarantee the global optimality of the solution. not seem, in principle, straightforward what might be the biolog-
ical plausibility of the optimal aeration profile for the first-stage
7.2.2. Optimisation results CS2 reactor, which remains nearly constant throughout the batch with
The search interval for the final time was [0.5, 1.5], and for the the exception of the spikes. However, in the case of the second
aeration flowrate [0.8, 15]. Again, the boundaries of the aeration stage (and excluding the spikes), the aeration flowrate is higher
flowrate have been chosen so as to allow wide variations, much at the beginning and lower at the end of the batch. This is the
higher than those seen in the actual plant (i.e., [3, 4]). The lower same qualitative solution that was found for CS1. Nevertheless, it is
boundary was also kept at 0.8 vvh, as aeration significantly con- not clear that the same biological plausibility applies here, because
tributes to sludge mixing in this kind of design. In this design, the second-stage reactor has a very low concentration of substrate
the first-stage reactor has an aeration power of 11 kW at a con- throughout the reaction. Again, it is unclear what may have lead the
stant aeration level of 4 vvh and the second-stage reactor 9 kW at algorithm to chose values pertaining to the “spiky” regions of the
3 vvh. Thus, the power-aeration relation over the aeration interval aeration flowrate. Generally, the algorithm did not obtain smooth
[0.8, 15] was assumed to be P1 = (11/4) · qa,1 for the first reactor and trajectories for the aeration flowrate. It can be observed that in
P2 = (9/3) · qa,2 for the second, where P is the power (kW) at the aer- this case lower aeration flowrates, lower polymer concentrations,
ation level qa (vvh). The polymer concentration (g/l) was also used and higher reaction times were favoured to minimise the energy
62 J. Rojas, T. Zhelev / Computers and Chemical Engineering 38 (2012) 52–63
Table 3 reactors was found to be lower than currently used. This means
Values of energy requirement, final time, and polymer concentration, prior and after
that it would be enough to keep it at about 2–3 vvh for one day to
optimisation for the two case-study plants.
satisfy treatment objectives.
Parameter CS1 CS2 Current ATAD models, including the one presented here, are
Base energy value (kWh/kg) 0.72 1.22 still in a relatively early stage of development when compared to
Minimum energy value (kWh/kg) 0.56 0.99 ASM models. To improve the accuracy of these models and that
Energy saving (%) 22 18 of optimisation results, further assessment, parameter estimation,
Base final time (days) 1 1
and model validation are needed. For this purpose, more quantity
Optimal final time (days) 0.7 1.08
Base polymer concentration (g/l) – 25 and higher quality full-scale data, at present extremely scarce, have
Opt. polymer concentration (g/l) – 6.5 to be made available.
Despite its advantages and common use in other fields, opti-
misation is still relatively rare in wastewater engineering. Given
requirement. It is also remarkable that the OUR of the first-stage the high, rising cost of wastewater treatment, optimisation should
reactor presents in this case a certain delay with respect to changes become the norm when it comes to design and operation of
in the aeration flowrate. wastewater treatment plants.
The optimal polymer concentration is 6.5 g/l and the total VS
concentration in the influent sludge, 21.5 g/l. This value goes is
lower than the minimum 25 g/l recommended by USEPA (1990) Acknowledgements
to consistently operate at thermophilic temperatures. However, as
shown by Fig. 6c the second-stage reactor consistently operates We wish to thank Jairo Gomez (NILSA, Navarra, Spain) for pro-
in the thermophilic temperature range. This result shows that the viding data and for his continued support. This publication has
assumption that VS concentration must be over 25 g/l to reach and emanated from research conducted with the financial support of
maintain thermophilic operation does not generally hold. Science Foundation Ireland under Grant No. 06/CP/E007.
The reaction time was slightly higher (1.08 days) than in the
typical 1-day operation. We noticed that this was the case for
Appendix A.
all optimal solutions in this particular case study. The algorithm
tended to find solutions close to this value, even when the start-
A.1. Details on optimisation results of CS1 shown in Fig. 5
ing point was far away from it or when starting with a randomized
matrix of starting points. This is consistent with the suggestion of
The initial condition of the system corresponding to the solution
Scisson (2003) that some ATAD systems may require longer (than 1
displayed in Fig. 5 is as follows:
day) final times to guarantee compliance with stabilisation require-
ments. It is not clear why in this particular case the algorithm did
x0 = [3, 1.5, 3, 7.8, 0.47, 0, 0.83, 8, 0, 62.8, 300]
not seek to further shorten the final time.
Table 3 summarises base and minimum energy values, the base
Note that all concentrations (i.e. from x1 to x9 ) are in g/l. x10
and optimal final time, and the energy savings for each of the two
stands here for the temperature (◦ C) instead of the total enthalpy
case studies.
of the liquid phase, and x11 for the total volume (m3 ).
As for the inputs, the amount of sludge replaced after each batch
8. Conclusions was 30 m3 (10% of total working volume). The total VS concentra-
tion in the influent sludge was 36.9 g/l, that is, the same as in the
The aim of this paper was to minimise the energy requirement of actual plant. The distribution of mass among the different compo-
ATAD while complying with treatment objectives, i.e. sludge stabil- nents in the influent sludge was as follows: x1 = 2.7 g/l, x2 = 4.5 g/l,
isation and pasteurisation. To this purpose, a dynamic ATAD model x3 = 2.7 g/l, x4 = 22.5 g/l, x5 = 3.1 g/l, x6 = 0.1 g/l, x7 = 1.35 g/l, x8 = 0 g/l,
was presented and assessed. A global sensitivity analysis was per- x9 = 0.6 mg/l, x10 = 10 ◦ C. The aeration flowrate is displayed in
formed to identify promising optimisation variables, that is, those Fig. 5d.
with the strongest influence on the energy requirement. It was
found that the aeration flowrate, the reaction time, and the sludge
flowrate (percentage of reactor volume replaced, loading time, and A.2. Details on optimisation results of CS2 shown in Fig. 6
reaction time) had a clear impact on the energy requirement, while
the influent temperature did not. As for the optimisation problem, The initial condition of the system corresponding to the solution
it was formulated following the sequential approach for dynamic displayed in Fig. 6 is as follows:
optimisation, due to the discontinuous, semi-batch nature of ATAD.
Two full-scale ATAD facilities were selected as case studies: a one- x01 = [0, 4.2, 0, 10.9, 0.01, 0, 1.82, 2.1, 0, 32.7, 100]
stage and a two-stage system. For the single-stage system, a 22%
improvement of the energy requirement was reached after opti-
misation, when compared to current plant operation. Similarly, for x02 = [0, 0.75, 0, 7.3, 0, 0, 1, 5.7, 0, 54.9, 100]
the two-stage system an improvement of 18% was attained. The
objective function was found to be highly multimodal in both case All concentrations (i.e. from x1 to x9 ) are in g/l. x10 stands here
studies, and the global optimality of the solutions can thus not be for the temperature (◦ C), and x11 for the total volume (m3 ).
guaranteed. As for the inputs, the amount of sludge replaced after each batch
Based on the optimisation results, the following recommen- was 25 m3 (12.5% of total working volume). The total VS concen-
dations can be made. For the single-stage system, the qualitative tration in the influent sludge was 21.5 g/l, that is, about the half
solution is obtained by applying a higher aeration flowrate at the as in the actual plant. The distribution of mass among the differ-
beginning of the reaction and decreasing it gradually until reaching ent components in the influent sludge was as follows: x1 = 0 g/l,
the lower boundary at the end of the batch. The final time should x2 = 9.2 g/l, x3 = 0 g/l, x4 = 11.7 g/l, x5 = 0 g/l, x6 = 0.1 g/l, x7 = 0.5 g/l,
be kept at its lowest value, corresponding to the highest sludge x8 = 0 g/l, x9 = 0.6 mg/l, x10 = 10 ◦ C. The aeration flowrate is displayed
flowrate. For the two-stage system, the aeration level in both in Fig. 6d.
References Kelly, H. G., & Mavinic, D. S. (2003). Autothermal thermophilic aerobic digestion
research, application and operational experience. In WEFTEC 2003 workshop
Banga, J. R., Balsa-Canto, E., & Moles, C. G. (2003). Dynamic optimization of biore- W104 Los Angeles, CA, October 2003.
actors: A review. Proceedings of the Indian National Science Academy, 69A, Kovács, R., Csikor, Zs., Házi, F., & Miháltz, P. (2007). Application of activated sludge
257–265. model no. 3 for the modeling of organic matter biodegradation at thermophilic
Banga, J. R., Balsa-Canto, E., Moles, C. G., & Alonso, A. A. (2005). Dynamic opti- temperatures. Water Environment Research, 79(5), 554–560.
mization of bioprocesses: Efficient and robust numerical strategies. Journal of Kovács, R., Miháltz, P., & Csikor, Zs. (2007). Kinetics of autothermal thermophilic
Biotechnology, 117(4), 407–419. aerobic digestion: Application and extension of activated sludge model no. 1 at
Barnes, D., Bliss, P. J., Gould, B. W., & Vallentine, H. R. (1986). Water and wastewater thermophilic temperatures. Water Science & Technology, 56(9), 137–145.
engineering systems. Longman Scientific & Technical. LaPara, T. M., & Alleman, J. E. (1999). Thermophilic aerobic biological wastewater
Batts, C. W., Burton, F. L., & Jones, M. (1993). Demand side management opportunities treatment. Water Research, 33(4), 895–908.
in the wastewater industry. In Water environment federation, proceedings of the Layden, N. M., Kelly, H. G., Mavinic, D. S., Moles, R., & Barlet, J. (2007a). Autother-
66th annual conference & exposition. mal thermophilic aerobic digestion (ATAD) – Part I: Review of origins, design,
Biegler, L. T. (2007). An overview of simultaneous strategies for dynamic optimiza- and process operation. Journal of Environmental Engineering and Science, 6(6),
tion. Chemical Engineering and Processing, 46, 1043–1053. 665–678.
Biegler, L. T. (2010). Nonlinear programming: Concepts algorithms and applications to Layden, N. M., Kelly, H. G., Mavinic, D. S., Moles, R., & Barlet, J. (2007b). Autothermal
chemical processes. SIAM—Society for Industrial and Applied Mathematics. thermophilic aerobic digestion (ATAD) – Part II: Review of research and full-
Bonvin, D., Palanki, S., & Srinivasan, B. (2003). Dynamic optimization of batch pro- scale operating experiences. Journal of Environmental Engineering and Science,
cesses: I. Characterization of the nominal solution. Computers and Chemical 6(6), 679–690.
Engineering, 27, 1–26. Le, S. M. (2006). Thermophilic biological pre-treatments for MADs. In AquaEnviro
Burton, F. L. (1996). Water and wastewater industries: Characteristics and energy workshop: Advances in technology for the anaerobic digestion of municipal sludge
management opportunities, CR-10691. St. Louis, MO: Electric Power Research Manchester, UK, 14 June 2006.
Institute. M/J Industrial Solutions. (2003). Municipal wastewater treatment plant energy base-
Commission of the European Communities. (2006). Action plan for energy effi- line study. Prepared for Pacific Gas and Electric Company, San Francisco CA.
ciency: Realising the potential. Technical Report COM(2006)545 final. Brussels: Metcalf and Eddy, Inc. (2003). Wastewater engineering: Treatment and reuse (4th ed.).
European Commission. New York: McGraw-Hill.
Csikor, Z., Mihaltz, P., Hanifa, A., Kovacs, R., & Dahab, M. F. (2002). Identification of fac- Moles, C. G., Gutierrez, G., Alonso, A. A., & Banga, J. R. (2003). Integrated process
tors contributing to degradation in autothermal thermophilic sludge digestion. design and control via global optimization: A wastewater treatment plant case
Water Science and Technology, 46(10), 131–138. study. Chemical Engineering Research and Design, 81, 507–517.
Descoins, N., Stephane, D., Remi, L., & Marechal, F. (2010). Energetic efficiency in Olsson, G., & Newell, B. (1999). Wastewater treatment systems. Modelling, diagnosis
Waste Water Treatments Plants: Optimization of activated sludge process cou- and control. London, UK: IWA Publishing.
pled with anaerobic digestion. In Proc. of the 23rd international conference on Ponti, C., Sonnleitner, B., & Fiechter, A. (1995a). Aerobic thermophilic treatment of
efficiency, cost, optimization, simulation and environmental impact of energy sys- sewage sludge at pilot plant scale. 1: Operating conditions. Journal of Biotech-
tems Lausanne, 14–17 June, (pp. 1062–1069). nology, 38(2), 173–182.
Egea, J. A., Balsa-Canto, E., García, M. G., & Banga, J. R. (2009). Dynamic optimization Ponti, C., Sonnleitner, B., & Fiechter, A. (1995b). Aerobic thermophilic treatment of
of nonlinear processes with an enhanced scatter search method. Industrial and sewage sludge at pilot plant scale. 2: Technical solutions and process design.
Engineering Chemistry Research, 48(9), 4388–4401. Journal of Biotechnology, 38(2), 183–192.
Egea, J. A., Martí, R., & Banga, J. R. (2010). An evolutionary method for Pontryagin, L. S. (1964). Mathematische Theorie Optimaler Prozesse. Wien: Olden-
complex-process optimization? Computers and Operations Research, 37(2), bourg Verlag.
315–324. Rojas, J. (2011). Energy efficiency optimisation of wastewater treatment: Study of
Fikar, M., Chachuat, B., & Latifi, M. A. (2005). Optimal operation of alternating acti- ATAD. Ph.D. Thesis, University of Limerick, Limerick, Ireland.
vated sludge processes. Control Engineering Practice, 13(7), 853–861. Rojas, J., Zhelev, T., & Bojarski, A. D. (2010). Modelling and sensitivity analysis of
Focus on Energy. (2006). Water and wastewater energy best practice guidebook. Pre- ATAD. Computers and Chemical Engineering, 34(5), 802–811.
pared for Wisconsin Department of Administration by the Focus on Energy Scisson, J. P. (2003). ATAD, the next generation: Design, construction, start-up and
Program. Madison, WI, US. operation of the first municipal 2nd generation ATAD. In Joint residuals and
Gomez, J. M. (2007). Digestion Aerobica Termofila Autosostenida (atad) de fangos: biosolids management conference and exhibition 2003.
Estudio experimental a escala real y modelizacion matematica del reactor. Ph.D. Toledo, R. T. (1991). Fundamentals of food process engineering (2nd ed.). New York:
Thesis, Universidad de Navarra, San Sebastian, Spain. Chapman and Hall.
Gomez, J., de-Gracia, M., Ayesa, E., & Garcia-Heras, J. L. (2007). Mathematical mod- USEPA. (1990). Environmental regulations and technology: Autothermal ther-
elling of autothermal thermophilic aerobic digesters. Water Research, 41(5), mophilic aerobic digestion of municipal wastewater sludge. Technical report,
959–968. United States Environmental Protection Agency, Office of Research and Devel-
Henze, M., Gujer, W., Mino, T., & van Loosdrecht, M. C. M. (2000). Activated sludge opment, Washington, DC.
models ASM1, ASM2, ASM2d and ASM3. Scientific and Technical Report No. 9. USEPA. (1993). 40 CFR PART 503. Standards for the use or disposal of sewage
London: IWA Publishing. sludge. Washington, DC: United States Environmental Protection Agency, Office
Holenda, B., Domokos, E., Redey, A., & Fazakas, J. (2007). Aeration optimization of of Research and Development.
a wastewater treatment plant using genetic algorithm. Optimal Control Applica- Warakomski, A., McWhirter, J., Balan, P., Swope, B., & Gyger, R. (2007). Evolution of
tions and Methods, 28, 191–208. atad/dual digestion and recent process developments. In IWA specialist confer-
Jones, M. (1991). Demand side management: Opportunities in water and sewage ence: moving forward wastewater biosolids sustainability: Technical, managerial,
treatment. IEEE Power Engineering Review, 11(11), 8–9. and public synergy Moncton, New Brunswick, Canada, June 2007.
Juteau, P. (2006). Review of the use of aerobic thermophilic bioprocesses for the Yonkin, M., Clubine, K., & O’Connor, K. (2008). Importance of energy efficiency to the
treatment of swine waste. Livestock Science, 102(3), 187–196. water and wastewater sector. Clear Waters, 38, 12–13.
Computers and Chemical Engineering 38 (2012) 52–63
Contents lists available at SciVerse ScienceDirect
Computers and Chemical Engineering

journal homepage: www.elsevier.com/locate/compchemeng
Energy efficiency optimisation of wastewater treatment: Study of ATAD

Jaime Rojas ∗ , Toshko Zhelev
Charles Parsons Initiative on Energy and Sustainable Environment, University of Limerick, National Technological Park, Plassey, Co. Limerick, Ireland
a r t i c l e i n f o a b s t r a c t
Article history: The aim of this paper is to minimise the energy requirement of autothermal thermophilic aerobic diges-
Received 4 May 2011 tion (ATAD). To this end, a dynamic ATAD model is presented and assessed. A global sensitivity analysis
Received in revised form was performed to identify the operating conditions with the strongest impact on the energy require-
13 November 2011
ments, and thus to choose the most promising optimisation variables. The latter turned out to be the
Accepted 29 November 2011
aeration flowrate, the reaction time, and the sludge flowrate. The optimisation problem was formu-
Available online 8 December 2011
lated following the sequential approach for dynamic optimisation, due to the discontinuous nature of
ATAD. The problem was implemented in MATLAB® and solved for two case studies using the eSS algo-
Keywords:
Wastewater treatment
rithm, a global scatter search method that alternates with local algorithms (in our case fmincon) to refine
ATAD the best solutions. The two selected full-scale case studies include a single-stage and a two-stage sys-
Energy efficiency tem. For the former, a 22% improvement of the energy requirement was achieved after optimisation,
Dynamic optimisation and 18% for the latter. Despite its advantages and common use in other fields, optimisation is still rel-
Sensitivity analysis atively rare in wastewater engineering. In the light of the high, rising cost of wastewater treatment,
optimisation should become the norm when it comes to design and operation of wastewater treatment
plants.
© 2011 Elsevier Ltd. All rights reserved.
1. Introduction 2% of the total amount of electricity generated (Batts, Burton, &

Jones, 1993). In fact, WWTPs represent the single largest cost to
1.1. Energy efficiency of wastewater treatment local governments with up to 33% of their total budget (Jones, 1991;
M/J Industrial Solutions, 2003; Yonkin, Clubine, & O’Connor, 2008),
In the present economic crisis with volatile and rising oil and their energy consumption is expected to increase by 30–40%
and energy prices, the steady increase of greenhouse gas emis- in the next 20–30 years (Metcalf and Eddy, 2003).
sions, and the foreseeable depletion of non-renewable energy In the light of these facts, it is surprising that there are very
sources, it has become evident that more sustainable energy few articles in the available literature devoted to the energy effi-
forms and measures are imperative to ensure a sustainable econ- ciency optimisation of wastewater treatment processes (Descoins,
omy for the future. One of those measures is energy efficiency Stephane, Remi, & Marechal, 2010; Fikar, Chachuat, & Latifi, 2005;
improvement (Commission of the European Communities, 2006). Holenda, Domokos, Redey, & Fazakas, 2007; Moles, Gutierrez,
It helps reducing energy requirements often with low investment Alonso, & Banga, 2003, and references cited therein).
cost. Descoins et al. (2010) pointed out that so far the wastewater
The principal concern of the wastewater industry has always modelling experts have focused on modelling the effluent wastew-
been to meet water quality standards in order to keep public ater qualities, while the implicit energy aspects have received very
trust (Focus on Energy, 2006). Thus, wastewater treatment plants little attention.
(WWTPs) are usually designed to meet certain effluent require- After labour, electricity is the largest operating cost associated
ments, without major energy considerations. As a result, WWTPs with wastewater treatment with 25–40% of the total (M/J Industrial
are hardly ever designed with energy efficiency in mind. Their Solutions, 2003). In the most common type of WWTP, the activated
design and operation is often based on intuition and experience sludge plant, about 50% of this energy is used for aeration purposes
(Metcalf and Eddy, 2003), rather than on optimal trajectories or set (Metcalf and Eddy, 2003).
points. Among the different treatment stages in a WWTP, sludge treat-
WWTPs are generally very energy-intensive and expensive to ment generally accounts for a major fraction of the total cost
operate. As an example, in the US alone WWTPs consume about (Barnes, Bliss, Gould, & Vallentine, 1986). There are mainly three
methods for sludge treatment: digestion (aerobic or anaerobic,
mesophilic or thermophilic), composting, and incineration. In this
∗ Corresponding author. paper, we will focus on a digestion process known as autothermal
E-mail addresses: [email protected], [email protected] (J. Rojas). thermophilic aerobic digestion (ATAD).
0098-1354/$ – see front matter © 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2011.11.016
applied
sciences
Article
Optimal Control of Wastewater Treatment Plants
Using Economic-Oriented Model Predictive
Dynamic Strategies
Silvana Revollar 1, * ID
, Pastora Vega 1 , Ramón Vilanova 2 and Mario Francisco 1
1 Department of Computing and automation, University of Salamanca, 37008 Salamanca, Spain;
[email protected] (P.V.); [email protected] (M.F.)
2 Department of Telecommunications and Systems Engineering School of Engineering,
Universitat Autònoma de Barcelona, 08193 Catalonia, Spain; [email protected]
* Correspondence: [email protected]; Tel.: +34-923-294-653
Received: 1 July 2017; Accepted: 1 August 2017; Published: 9 August 2017
Abstract: This paper addresses the implementation of economic-oriented model predictive controllers
for the dynamic real-time optimization of the operation of wastewater treatment plants (WWTP).
Both the economic-optimizing controller (pure-EMPC) and the economic-oriented tracking controller
(Hybrid-EMPC, or HEMPC) formulations are validated in the benchmark simulation model (BSM1)
platform that represents the behavior of a characteristic activated sludge process. The objective
of the controllers is to ensure the appropriate operation of the plant, while minimizing the energy
consumption and the fines for violations of the limits of the ammonia concentration in the effluent
along the full operating period. A non-linear reduced model of the activated sludge process is used
for predictions to obtain a reasonable computing effort, and techniques to deal with model-plant
mismatch are incorporated in the controller algorithm. Different designs and structures are compared
in terms of process performance and energy costs, which show that the implementation of the
proposed control technique can produce significant economic and environmental benefits, depending
on the desired performance criteria.
Keywords: wastewater treatment plant; activated sludge process; dynamic optimization; economic
model predictive control
1. Introduction
In the management of a wastewater treatment plant (WWTP) operation, one of the most important
factors that determines economics is the energy used to provide oxygen to the aerobic processes
(aeration energy), and the pumping energy for the recycles of the plant. The appropriate adjustment of
the available manipulated variables is crucial for an optimum operation, especially in the activated
sludge process, where biological removal of nutrients and organic matter takes place [1–3].
The influent of the wastewater treatment plants exhibits an oscillating behavior, with daily
and seasonal patterns associated with the human activities during the day and the seasonal rainfall.
Weather conditions, such as rain and storms, produce significant changes in the influent flowrate
and load [4]. Due to the variable influent behavior, the pollution load to be treated is continuously
changing, and consequently, so are the energy and chemical requirements for the treatment. In such a
scenario, conservative operation regulating the critical variables around the nominal working point
might ignore the disturbances introduced by the influent. It is forecasted that significant energy
savings could be achieved with an operation based on dynamic optimization that accounts for the
influent conditions.
Appl. Sci. 2017, 7, 813; doi:10.3390/app7080813 www.mdpi.com/journal/applsci

Appl. Sci. 2017, 7, 813 13 of 21
Appl. Sci. 2017, 7, 813 13 of 21
Appl. Sci. 2017, 7, 813 13 of 21
Appl. Sci. 2017, 7, 813 13 of 21
(a)
(a) (b)
(b)
(a) (b)
Figure 5. Control signals applied to the BSM1 plant operating under NMPC and HEMPC controllers.
Figure 5. Control signals applied to the BSM1 plant plant operating
operating under
under NMPC
NMPC and
and HEMPC
HEMPC controllers.
controllers.
Figure
(a) 5. Control
Internal recyclesignals
flow applied
(Qa) (b) to the BSM1
oxygen plant
transfer operating
coefficient under NMPC and HEMPC controllers.
(KLa).
flow (Qa)
(a) Internal recycle flow (Qa) (b)
(b) oxygen
oxygen transfer
transfer coefficient
coefficient (KLa).
(KLa).
(a) Internal recycle flow (Qa) (b) oxygen transfer coefficient (KLa).
(a)
(a) (b)
(b)
(a) (b)
Figure 6.
Figure 6. BSM1-controlled
6. BSM1-controlled variables’
variables’responses
responses under NMPC
under NMPC and
NMPC and EMPC
and EMPC controllers.
controllers.(a)
EMPC controllers. Nitrates
(a) Nitrates and
and
Figure
Figure
nitrites 6. BSM1-controlled
BSM1-controlled
concentration ininthe variables’
variables’
second responses
responses
rector (S(SNO2under
under NMPC and EMPC controllers. (a)
(a) Nitrates
Nitrates
) response (b) dissolved oxygen in the last aerobic and
and
nitrites concentration
nitrites concentration
concentration in the
in the second
the second rector
second rector
rector (S NO2) response (b) dissolved oxygen in the last aerobic
(SNO2 )response
response(b)
(b)dissolved
dissolvedoxygen
oxygeninin the
the last
last aerobic
aerobic
nitrites
reactor NO2)model
reactor(S
reactor (SO5O5) )response.
(S ) response.
response.
EMPC:
EMPC:
EMPC:
economic-oriented
economic-oriented
economic-oriented model
model
predictive
predictivecontroller.
predictive controller.
controller.
reactor (SO5) response. EMPC: economic-oriented model predictive controller.
O5
(a)
(a) (b)
(b)
(a) (b)
Figure
Figure 7. Control signals applied to the BSM1 plant operating under NMPC andEMPC
7. Control signals applied to the BSM1 plant operating under NMPC and EMPCcontrollers.
controllers.
Figure
(a)
Figure 7.
Internal
7. Control
recycle
Control signals
flow
signals applied
(Qa) (b)
applied to
oxygen
to the
the BSM1
transfer
BSM1 plant operating
coefficient
plant under
(KLa).
operating under NMPC
NMPC and
and EMPC
EMPC controllers.
controllers.
(a) Internal recycle flow (Qa) (b) oxygen transfer coefficient (KLa).
(a)
(a) Internal
Internal recycle
recycle flow
flow (Qa)
(Qa) (b)
(b) oxygen
oxygen transfer
transfer coefficient (KLa).
coefficient (KLa).
Appl. Sci. 2017, 7, 813 14 of 21
Appl. Sci. 2017, 7, 813 14 of 21
Appl. Sci. 2017, 7, 813 14 of 21
(a) (b)
Figure 8. Instantaneous pumping (a) and aeration energy (b) for the BSM1 plant operating under
selected controllers.
selected controllers.
The performance indices calculated for the overall operation period with the proposed
controllers are presented
The performance in Tables
indices calculated 2 andfor 3, the
respectively. The influent
overall operation profile
period(b)with the described
proposed in controllers
the BSM1
(a)
for
aredry weatherin(Figure
presented Tables2)2 isand used 3, to evaluate theThe
respectively. plant performance
influent profileover an operation
described in the period
BSM1 for of four
dry
days,
weatherthe (Figure
first selected
day2)withis constant
used
controllers. to influent.
evaluate The
the influent
plant quality
performance index
over defined
an in
operation [22] to measure
period of the
four
pollution load
days, the firstThe of the influent
dayperformance
with constantis IQ = 55,945.14 kg-pollutants/day. The indices
influent. The influent quality index defined in [22] to measure computed for the default
indices calculated for the overall operation period with the proposed
BSM1 strategy
the pollution are
load included
of the also
influent
controllers are presented in Tables for
is IQcomparison.
2=and
55,945.14 TableThe 2 contains
kg-pollutants/day.
3, respectively. theThe
influent profile indices
indices
described associated
the BSM1 with
in computed for the
the
variables considered weatherin the economic-oriented
(Figure 2) is used to evaluate model
the plantpredictive
default BSM1 strategy are included also for comparison. Table 2 contains the indices associated with
for dry performance controller
over an optimization,
operation period of to
four show that
performance
the variables days, the first day
indices
considered vary with
in constant
according
the influent.
to theTherelative
economic-oriented influent
model quality
importanceindex defined
predictive given in to
controller [22]each
to measure the toinshow
objective
optimization, the
pollution load of the influent is IQ = 55,945.14 kg-pollutants/day. The indices computed for the default
optimization.
that performance In general,
indices allvary
the EMPC according designs to improve
the relativethe importance
pumping energy given with
to respect
each to the NMPC
objective in the
BSM1 strategy are included also for comparison. Table 2 contains the indices associated with the
and PI default
optimization. strategies
In general,
variables based
consideredall on set-point
theeconomic-oriented
in the EMPC designs tracking.
improve
model A reduction
predictive in
thecontroller
pumping the energy
energy to
optimization, costs
with (AE
showrespect
that + PE) of
to the
13%
NMPC withand respect to the
PI default
performance NMPC varyis
strategies
indices achieved,
based
according to with
on set-point the pure
tracking.
the relative EMPC focused
A reduction
importance given to in onthethe
each optimization
energy
objective in the (AEof
costs + the
PE)
energy
of 13% costs
with optimization.
(EMPC-OCI).
respect to In the
general,
NMPC all the is
Ammonium EMPC designs improve
concentration
achieved, with the thethe
in pumping
pure effluent
EMPCenergy with respect
reduces
focused to
byon21% the
the NMPCthe EMPC
when
optimization of
and PI default strategies based on set-point tracking. A reduction in the energy costs (AE + PE) of
is focused
the energy13% on
coststhe minimization of the off-specification ammonium
(EMPC-OCI). Ammonium concentration in the effluent reduces by 21% when the concentration in the effluent.
with respect to the NMPC is achieved, with the pure EMPC focused on the optimization of the
Comparing
EMPC is focused with the default
on the
energy costs BSM1
minimization
(EMPC-OCI). PI strategy,
Ammonium of the a reduction
in the of
off-specification
concentration the energy
ammonium
effluent reduces bycosts21%of
concentration
when13%theisEMPC
achieved
in also
the effluent.
with the EMPC-OCI, produced especially by a decrease in the pumping
Comparing with the default BSM1 PI strategy, a reduction of the energy costs of 13% is achieved also
is focused on the minimization of the off-specification ammonium energy
concentration inof up
the to 23%.
effluent.
Comparing with
with the EMPC-OCI, the default
produced BSM1 PI strategy,
especially a reduction
by a decrease inofthe
the pumping
energy costsenergy
of 13% isof achieved
up to also
23%.
with the EMPC-OCI, produced especially by a decrease in the pumping energy of up to 23%.
Figure 9. Ammonium concentration in the effluent for selected controller designs.

Figure 9. Ammonium concentration in the effluent for selected controller designs.
Table 3 is presented to show that additional process variables are affected by the economic
Figure 9. and
optimization, Ammonium concentration
it is possible ininthe
to include them theeffluent for selected
optimization. The OCIcontroller designs.
index reported in Table 3
Table 3 is presented to show that additional process variables are affected by the economic
optimization,
Table 3 isand it is possible
presented to include
to show them in theprocess
that additional optimization. OCIaffected
Theare
variables index reported in Table 3
by the economic
optimization, and it is possible to include them in the optimization. The OCI index reported in Table 3
Appl. Sci. 2017, 7, 813 15 of 21
corresponds to the BSM1 performance index, which includes aeration energy, pumping energy, mixing
energy and sludge production costs.
Table 2. Performance indices associated with variables of the plant optimized in the economic oriented
model predictive controllers cost function (computed for a four-day operation period under dry
weather disturbances). NMPC: non-linear model predictive controller; HEMPC: economic-oriented
tracking controller; OCI: overall cost index penalization; NH: off-specification ammonium load in the
effluent penalization; EMPC: economic-oriented model predictive controller.
Average SNH in
EQ AE PE PE + AE SNH Violations
Controller the Effluent
(kg/Day) (kWh/Day) (kWh/Day) (EUR/Day) (Days)
(mg/L)
Default PI 2.79 6587.7 3785 268.0 4053.0 0.84
NMPC 3.00 6573.8 3802 276.6 4078.6 0.96
HEMPC-OCI 3.21 6783.9 3796 252.8 4048.8 1.00
HEMPC-OCI + NH 2.74 6584.7 3830 246.8 4076.8 0.88
HEMPC-NH 2.55 6698.0 3855 224.8 4079.8 0.73
EMPC-OCI 6.05 8289.0 3310 206.0 3516.0 3.34
EMPC-OCI + NH 3.47 7617.0 3841 167.2 4008.2 1.15
EMPC-NH 2.37 7315.3 4277 166.0 4443.0 0.74
In order to compare with the previous work of Zeng and Liu [16], Table 4 presents the changes
in the performance indices, with respect to the operation with the BSM1 default PI strategy in the
same simulation scenario. The comparison is made with selected designs with analogous tuning
parameters. The EMPC proposed in this work enhances the economic performance mainly by reducing
the pumping energy in all the cases, and the aeration energy in the EMPC-OCI design, which are the
operation costs considered in the economic function. In the Zeng and Liu [16] formulation, the BSM1
overall cost index (OCI) is optimized; it includes aeration and pumping energy, but also the mixing
energy and the costs of sludge disposal. Then, an operation with small energy consumption with
respect to the default PI strategy is achieved in general, but pumping aeration costs are larger than
those obtained with the methodology proposed here. On the other hand, since the effluent quality
index is included as an optimization objective in the Zeng and Liu [16] formulation, the improvement
in this index is notorious, while the advantages of the EMPC formulation proposed in this paper are
associated with the economic performance. In the current work, it has been preferred to not work
directly with the overall effluent quality index as part of the cost function, because it involves several
concentrations not available for measurement. Instead, here, it has been preferred to keep the two main
variables of interest (oxygen and nitrate) around desired values, while attempting to keep effluent
ammonia under the established limits.
Table 3. Performance indices of the mean operating variables of the plant that are different from the
variables included in the optimization problem (computed for a four-day operation period under dry
weather disturbances).
Average SNH in
TN Sludge Produced OCI SNH Violations TN Violations
Controller the Effluent
(mg/L) (kg/Day) (EUR/Day) (Days) (Days)
(mg/L)
Default PI 2.79 17.2 2747.0 18,031 0.84 0.98
NMPC 3.00 16.7 2747.3 18,055 0.96 0.73
HEMPC-OCI 3.21 17.4 2746.6 18,021 1.00 0.94
HEMPC-OCI + NH 2.74 17.3 2747.0 18,052 0.88 0.87
HEMPC-NH 2.55 18.2 2748.0 18,064 0.73 1.47
EMPC-OCI 6.05 19.5 2741.9 17,466 3.34 3.60
EMPC-OCI + NH 3.47 21.2 2748.0 17,988 1.15 3.80
EMPC-NH 2.37 21.8 2750.0 18,434 0.74 3.95
Regarding this reduction in energy consumption, as the work conducted here is based on the
BSM1 benchmark scenario, no specific WWTP was used. However, the BSM1 protocols define influent
Appl. Sci. 2017, 7, 813 16 of 21
characteristics featuring a load of around 100,000 person equivalents (PE) (80,000 from households,
and 20,000 from industrial origin). This helps to provide an idea of the repercussion of the mentioned
energy savings. Of course, in order to get concrete monetary figures of actual order, there will
be the need to link the cost indexes to actual energy costs. This may be highly dependent on the
WWTP location. However, just to point to some figures, according to HUBER Technology (http:
//www.picatech.ch/solutions/energy-efficiency), power consumption of state-of-the-art wastewater
treatment plants should be around 45 kWh/(PE.a) for plants serving >100,000 PE, where pumping and
aeration are recognized primary power consumers.
Table 4. Performance indices variation respect to the BSM1 PI default strategy.
Controller % AE Variation % PE Variation % OCI Variation

EMPC-OCI −12.55 −23.13 −3.13
EMPC-OCI + NH 1.47 −37.61 −0.24
EMPC-NH 13.00 −38.06 2.24
Zeng and Liu EMPC-NH 3.73 3.30 −9.78
Zeng and LiuEMPC-OCI + NH −3.40 6.75 −11.18
4. Conclusions
In this paper, a successful implementation of a single-layer economic oriented model predictive
control approach for the optimization of the operation of wastewater treatment plants (WWTPs) in a
simulation environment is presented. Even though this study used empirical models based on data
gathering and the adequate software and communication systems, the technique could be implemented
in real plants.
An OCI reduction of 13% with respect to the standard NMPC focused on set-point tracking is
achieved, with the pure EMPC focused on the optimization of the energy costs (EMPC-OCI), and
ammonium concentration in the effluent reduces by 21% when using the EMPC focused on the
minimization of the off-specification ammonium concentration in the effluent. Those improvements
are achieved by performing a dynamic optimization of the operation, taking advantage of the variations
in the load to find more favourable operating conditions on a given operating window.
The results demonstrate that the implementation of novel advanced control techniques as
single-layer EMPCs can produce significant energy savings and improvements in WWTP performance.
The single-layer strategy has the advantage of being simpler than other strategies, e.g., [9], which
obtain further reductions in costs but with a more complex hierarchical framework for nitrate and
oxygen control. On the other hand, in the pioneering implementation of EMPCs to the BSM1 model
presented in [16] the full model is used for predictions, therefore, a novel feature presented in this work
is the implementation of techniques to face modeling errors. It is also encouraging the potential use of
a reduced model with partial information with prediction purposes. Alternative formulations oriented
towards the increase of model quality should be investigated, as they will have a direct repercussion
on the final quality of the obtained control.
Future work is oriented to exploit these characteristics in different scenarios, and introduce
additional optimization objectives in the economic-oriented controller formulation. It is encouraging
to promote continuing research in the application of these advanced control strategies to improve
WWTP operation.
Acknowledgments: The authors wish to thank the support of the Spanish Government through the Ministerio
de Economía y Competitividad (MINECO) projects DPI2015-67341-C2-1-R, DPI2016-77271-R also with FEDER
funding. The Samuel Solórzano Foundation through project FS/21-2015 and the IWA Task Group from the
Department of Industrial Electrical Engineering and Automation (IEA), Lund University, Sweden (Ulf Jeppsson,
Christian Rosen) for the BSM1 models.
Author Contributions: Pastora Vega and Ramón Vilanova conceived and designed the experiments;
Silvana Revollar performed the experiments and Silvana Revollar and Mario Francisco analyzed the results; all the
authors have contributed to write the paper.
Appl. Sci. 2017, 7, 813 17 of 21
Conflicts of Interest: The authors declare no conflict of interest.
Appendix A
The BSM1 platform uses the Activated Sludge Process No1 (ASM1) to represent the behavior of the
biological reactors. It uses 13 state variables on each reactor, and considers eight biological processes such
as heterotrophic and autotrophic biomass growth in anoxic and aerobic conditions and biomass decay,
ammonification of soluble organic nitrogen, hydrolysis of entrapped organics, and hydrolysis of entrapped
organic nitrogen. In the reduced model used in this work, only the faster biological processes are taken
into account: the growth of heterotrophic biomass in aerobic conditions (ρ1i ), the growth of heterotrophic
biomass in anoxic conditions (ρ2i ), and the growth of autotrophic biomass in aerobic conditions (ρ3i ). The
plant is reduced to one anoxic reactor and one aerobic reactor, with a volume equivalent to the anoxic and
aerobic zones of the BSM1, respectively. The concentration of autotrophic and heterotrophic biomass in
the reactors is assumed to be constant (XB,H = 2500 mg/L, XB,A = 150 mg/L). These constant values are
taken from [17] and correspond to the steady state values at the biological reactors. The influent variables
considered in the reduced model are: the influent flow (Qin ), the organic matter concentration (SS,in ), and
the ammonium concentration (SNH,in ). The index i is used to distinguish the units, with reactor 1 the anoxic
zone (i = 1), and reactor 2 the aerobic zone (i = 2). The states associated to the faster biological processes are:
the concentration of ammonium compounds (SNHi ), the concentration of nitrates (SNOi ), the organic matter
concentration (SSi ) and the oxygen concentration (SOi ). The inputs are the oxygen transfer coefficient (KLa )
and the internal recycle flow (Qa ).
Regarding the comparison with real plants measurements, although the BSM1 model variables
and reduced model variables differ from the variables typically measured in real wastewater treatment
plants, it is possible to form a relationship between their values. Some equations to compute
COD (chemical oxygen demand), BOD (Biological oxygen demand) and other typical wastewater
measurements from the BSM1 variables are presented in Alex et al. [22].
A list of the process parameters and variables considered in the reduced model used here are
presented in Table A1.
Table A1. Reduced BSM1 model variables and parameters for a temperature close to T = 15 ◦ C.
Description Name Values

Process variables
Anoxic reactor volume (m3 ) V1 2000
Aerobic reactors volume (m3 ) V2 3999
Active heterotrophic biomass concentration (gr COD/m3 ) XB.H 2500
Active autotrophic biomass concentration (gr COD/m3 ) XB,A 150
Dissolved oxygen concentration (gr COD/m3 ) SO 0.1–5
Nitrate and nitrite concentration (gr N/m3 ) SNO 0.1–5
NH4 + + NH3 concentration (gr N/m3 ) SNH 0–14
Readily biodegradable substrate concentration (gr COD/m3 ) SS 0–40
Influent flow rate (m3 /h) Qin 45–2500
Effluent flow rate (m3 /h) Qe 45–2500
Readily biodegradable substrate concentration in the influent (gr COD/m3 ) SS,in 45–140
Ammonium compounds concentration in the influent (gr N/m3 ) SNH,in 8–55
Internal recycle flow (m3 /h) Qa 180–2000
Oxygen transfer coefficient (1/h) KLa 0–10
Biological processes parameters
Oxygen saturation concentration (gr/m3 ) SO,sat 8
Heterotrophic max. specific growth rate (1/day) µH 4
Half saturation coefficient for heterotrophs (gr COD/m3 ) Ks 10
Oxygen saturation coefficient for heterotrophs (gr COD/m3 ) KO,H 0.2
Ammonia saturation coefficient for heterotrophs (grN/m3 ) KNH 1
Oxygen saturation coefficient for autotrophs (gr COD/m3 ) KO,A 0.4
Heterotrophic yield (gr cell COD formed/gr COD oxydized) YH 0.67
Autotrophic yield (gr cell COD formed/gr N oxydized) YA 0.24
Nitrogen fraction in biomass (gr N/gr COD) iXB 0.08
Appl. Sci. 2017, 7, 813 18 of 21
The equations of the mathematical model are the following:

Biological processes:

SSi SOi
ρ1i = µ H · · XB,H (A1)
Ks + SSi KO,H + SOi

SSi KO,H SNOi
ρ2i = µ H · · n g XB,H (A2)
Ks + SSi KO,H + SOi KNO + SNOi

SNHi SOi
ρ3i = µ A · · XB,A (A3)
KNH + SNHi KO,A + SOi
Differential equations given by the mass balances:

dSNH1 1 1
dt = V1 [ Qin · SNHin + Q a SNH2 − ( Qin + Q a ) · SNH1 ] − i Xb · ρ11 − i Xb · ρ21 − i Xb + YA · ρ31 (A4)
dSNO1 1 1 − YH 1
= [ Q a SNO2 − ( Qin + Q a )SNO1 ] − ρ + ·ρ (A5)
dt V1 2.86YH 21 YA 31
dSS1 1 1 1
= [ Qin · SSin + Q a SS2 − ( Qin + Q a )SS1 ] − ρ11 − ρ (A6)
dt V1 YH YH 21
1 − YH

dSO1 1 4.57
= [ Q a · SO2 − ( Q0 + Q a )SO1 ] − ρ11 + + 1 ρ31 (A7)
dt V1 YH YA

dSNH2 1 1
= [( Qin + Q a )SNH1 − ( Qin + Q a )SNH2 ] − i xb · ρ12 − i xb + · ρ32 (A8)
dt V2 YA
dSNO2 1 1 − YH 1
= [( Qin + Q a )SNO1 − ( Qin + Q a )SNO2 ] − ρ22 + · ρ32 (A9)
dt V2 2.86YH YA
dSS2 1 1 1
= [( Qin + Q a )SS1 − ( Qin + Q a )SS2 ] − ρ12 − ρ22 (A10)
dt V2 YH YH
h i
dSO2 1 1−YH 4.57−YA
dt = V2 [( Qin + Q a ) · SO1 − ( Qin + Q a )SO2 ] − YH ρ12 + YA ρ32 + KLa(SO,sat − SO2 ) (A11)
Appendix B
The comparison between the reduced model responses and the BSM1 model responses of the
most relevant variables for the controller optimization, namely, nitrates concentration at the end of the
anoxic zone (SNO2 in BSM1, SNO1 in reduced model), oxygen concentration at the end of the aerobic
zone (SO5 in BSM1, SO2 in reduced model), and ammonium concentration at the end of the aerobic
zone (SNH5 in BSM1, SNH2 in reduced model), is presented in Figures A1–A3.
The model behavior is similar to BSM1 response in the evaluated conditions; however, a significant
difference in the magnitude of the variables is observed. However, in the evaluation of model for
predictions in the controller executions, the difference is lowered, since an initial state given by the
measured BSM1 states is provided.
It is important to highlight that the objective of the work is not the development of an accurate
reduced BSM1 model. The objective is to evaluate the performance of the single-layer EMPC for
optimizing economics in the presence of modeling errors.
The reduced model represents the variables with the faster dynamics, which are directly affected
by control actions, and are optimized each sampling time (5 min in this implementation). Even
though the hydrolysis of particulate matters is relevant in the biological processes that take place
in the activated sludge process, its dynamics can be neglected in these applications. A sensitivity
analysis of an accurately reduced model considering the different dynamics of the biological processes
is presented in [29].
Appl.
Appl. Sci. 2017, 7, 813 19 of 21
Appl.Sci. 2017,7,7,813
Sci.2017, 813 19
19ofof21
21
Figure A1. Comparison of the reduced model and BSM1 evolution of the nitrates concentration in the
Figure A1.Comparison
FigureA1. Comparisonof ofthe
thereduced
reducedmodel
modeland
andBSM1
BSM1evolution
evolutionof
ofthe
thenitrates
nitratesconcentration
concentrationininthe
the
second
second anoxic reactor
reactor obtainedininthe
theopen
open loop operation using
thethe manipulated variables values
secondanoxic
anoxic reactorobtained
obtained in the loop
open operation
loop using
operation using manipulated
the manipulatedvariables values
variables that
values
that produce
produce the nominal
the nominal set points.
set points.
that produce the nominal set points.
Figure A2.
Figure Comparison of
A2. Comparison of the
the reduced
reduced model
model and
and BSM1
BSM1 evolution
evolution ofof the
the oxygen
oxygen concentration
concentration in
in the
the
Figure A2. Comparison of the reduced model and BSM1 evolution of the oxygen concentration in the
fifth reactor (aerobic) obtained in the open loop operation, using the manipulated variables values
fifth reactor (aerobic) obtained in the open loop operation, using the manipulated variables values that
fifth reactor (aerobic) obtained in the open loop operation, using the manipulated variables values
produce
that the nominal
produce set points.
that produce the nominal set points.
Appl. Sci. 2017, 7, 813 20 of 21
Appl. Sci. 2017, 7, 813 20 of 21
Figure A3. Comparison of the reduced model and BSM1 evolution of the ammonia concentration in
Figure
the A3. obtained
effluent Comparison of open
in the the reduced model and
loop operation BSM1
using evolution of the
the manipulated ammonia
variables concentration
values that producein
the effluent obtained
the nominal set points.in the open loop operation using the manipulated variables values that produce
References
References
1. Olsson, G.; Nielsen, M.; Yuan, Z.; Lynggaard-Jensen, A.; Steyer, J.-P. Instrumentation, Control and Automation
1. Olsson, G.; Nielsen, M.; Yuan, Z.; Lynggaard-Jensen, A.; Steyer, J.-P. Instrumentation, Control and Automation
in Wastewater Systems; IWA Publishing: London, UK, 2005.
in Wastewater Systems; IWA Publishing: London, UK, 2005.
2. Olsson, G. ICA and me a subjective review. Water Res. 2012, 46, 1585–1624. [CrossRef] [PubMed]
2. Olsson, G. ICA and me a subjective review. Water Res. 2012, 46, 1585–1624.
3. Santin, I.; Pedret, C.; Vilanova, R. Control and Decision Strategies in Wastewater Treatment Plants for Operation
3. Santin, I.; Pedret, C.; Vilanova, R. Control and Decision Strategies in Wastewater Treatment Plants for Operation
Improvement; Springer: Cham, Switzerland, 2016.
Improvement; Springer: Cham, Switzerland, 2016.
4. Gernaey, K.; Rosen, C.; Jeppsson, U. BSM2: A Model for Dynamic Influent Data Generation; Technical Report;
4. Gernaey, K.; Rosen, C.; Jeppsson, U. BSM2: A Model for Dynamic Influent Data Generation; Technical Report;
Lund University: Lund, Sweden, 2005.
5. Amand, L.; Olsson, G.; Carlsson, B. Aeration Control—A review. Water Sci. Technol. 2013, 67, 2374–2397.
5. Amand, L.; Olsson, G.; Carlsson, B. Aeration Control—A review. Water Sci. Technol. 2013, 67, 2374–2397.
[CrossRef] [PubMed]
6. O’Brien, M.; Mack, J.; Lennox, B.; Lovett, D.; Wall, A. Model predictive control of an activated sludge
6. O’Brien, M.; Mack, J.; Lennox, B.; Lovett, D.; Wall, A. Model predictive control of an activated sludge process:
process: A case study. Control Eng. Pract. 2011, 19, 54–61.
A case study. Control Eng. Pract. 2011, 19, 54–61. [CrossRef]
7. Santín, I.; Pedret, C.; Vilanova, R. Fuzzy control and model predictive control configurations for effluent
7. Santín, I.; Pedret, C.; Vilanova, R. Fuzzy control and model predictive control configurations for effluent
violations removal in wastewater treatment plants. Ind. Eng. Chem. Res. 2015, 54, 2763–2775.
violations removal in wastewater treatment plants. Ind. Eng. Chem. Res. 2015, 54, 2763–2775. [CrossRef]
8. Vega, P.; Revollar, S.; Francisco, M.; Martin, J.M. Integration of set point optimization techniques into
8. Vega, P.; Revollar, S.; Francisco, M.; Martin, J.M. Integration of set point optimization techniques into
nonlinear MPC for improving the operation of WWTPs. Comput. Chem. Eng. 2014, 68, 78–95.
nonlinear MPC for improving the operation of WWTPs. Comput. Chem. Eng. 2014, 68, 78–95. [CrossRef]
9. Santín, I.; Pedret, C.; Vilanova, R. Applying variable dissolved oxygen set point in a two level hierarchical
9. Santín, I.; Pedret, C.; Vilanova, R. Applying variable dissolved oxygen set point in a two level hierarchical
control structure to a wastewater treatment process. J. Process Control 2015, 28, 40–55.
control structure to a wastewater treatment process. J. Process Control 2015, 28, 40–55. [CrossRef]
10. Santín, I.; Pedret, C.; Vilanova, R.; Meneses, M. Removing violations of the effluent pollution in a
10. Santín, I.; Pedret, C.; Vilanova, R.; Meneses, M. Removing violations of the effluent pollution in a wastewater
wastewater treatment process. Chem. Eng. J. 2015, 279, 207–219
treatment process. Chem. Eng. J. 2015, 279, 207–219. [CrossRef]
11. Piotrowski, R.; Brdys, M.A.; Konarczak, K.; Duzinkiewicz, K.; Chotkowski, W. Hierarchical dissolved
11. Piotrowski, R.; Brdys, M.A.; Konarczak, K.; Duzinkiewicz, K.; Chotkowski, W. Hierarchical dissolved oxygen
oxygen control for activated sludge processes. Control Eng. Pract. 2008, 16, 114–131.
control for activated sludge processes. Control Eng. Pract. 2008, 16, 114–131. [CrossRef]
12.
12. Brdys,M.A.;
Brdys, M.A.;Grochowski,
Grochowski,M.; M.; Gminski,
Gminski, T.; T.;Konarczak,
Konarczak,K.;
K.;Drewa,
Drewa,M. M.Hierarchical
Hierarchicalpredictive
predictivecontrol
controlofof
integrated wastewater treatment systems. Control Eng. Pract. 2008, 16, 751–767.
integrated wastewater treatment systems. Control Eng. Pract. 2008, 16, 751–767. [CrossRef]
13.
13. Darby,M.L.;
Darby, M.L.;Nikolau,
Nikolau,M.; M.;Jones,
Jones,J.;J.;Nicholson,
Nicholson,D.D.RTO:
RTO:AnAnoverview
overviewandandassessment
assessment of
of current
current practice.
practice.
J. Process Control. 2011, 21, 874–884.
J. Process Control. 2011, 21, 874–884. [CrossRef]
14.
14. Tatjewski,P.P.Advanced
Tatjewski, Advancedcontrol
controland
andon-line
on-lineprocess
processoptimization
optimizationininmultilayer
multilayerstructures.
structures.Ann.
Ann.Rev.
Rev.Control
Control
2008, 32, 71–85.
2008, 32, 71–85. [CrossRef]
15. Revollar, S.; Vega, P.; Vilanova, R.; Francisco, M.; Santín, I. Optimization of Economic and Environmental
Objectives in a Non Linear Model Predictive Control Applied to a Wastewater Treatment Plant.
In Proceedings of the 20th ICSTCC, Sinaia, Romania, 13–15 October 2016.
Appl. Sci. 2017, 7, 813 21 of 21
15. Revollar, S.; Vega, P.; Vilanova, R.; Francisco, M.; Santín, I. Optimization of Economic and Environmental
Objectives in a Non Linear Model Predictive Control Applied to a Wastewater Treatment Plant.
In Proceedings of the 20th ICSTCC, Sinaia, Romania, 13–15 October 2016.
16. Zeng, J.; Liu, J. Economic model predictive control of wastewater treatment processes. Ind. Eng. Chem. Res.
2015, 54, 5710–5721. [CrossRef]
17. Ellis, M.; Duranda, H.; Christofides, P. A tutorial review of economic model predictive control methods.
J. Process Control 2014, 24, 1156–1178. [CrossRef]
18. Würth, L.; Hannemann, R.; Marquardt, W. A two-layer architecture for economically optimal process control
and operation. J. Process Control 2011, 21, 309–436. [CrossRef]
19. Engell, S. Feedback control for optimal process operation. J. Process Control 2007, 17, 203–219. [CrossRef]
20. Forbes, J.F.; Marlin, T.E. Model accuracy for economic optimizing controllers: The bias update case. Ind. Eng.
Chem. Res. 1994, 33, 1919–1929. [CrossRef]
21. Samuelsson, P.; Halvarsson, B.; Carlsson, B. Cost-efficient operation of a denitrifying activated sludge process.
Water Res. 2007, 41, 2325–2332. [CrossRef] [PubMed]
22. Alex, J.; Benedetti, L.; Copp, J.; Gernaey, K.; Jeppsson, U.; Nopens, I.; Pons, M.; Rieger, L.; Rosen, C.; Steyer, J.;
et al. Benchmark Simulation Model No. 1 (BSM1), IWA Taskgroup on Benchmarking of Control Strategies for
WWTPs; Cod.: LUTEDX-TEIE 7229. 1-62; Department of Industrial Electrical Engineering and Automation,
23. Maciejowski, J.M. Predictive Control with Constraints; Prentice Hall: Upper Saddle River, NJ, USA, 2002.
24. Qin, S.; Badgwell, T. A survey of industrial model predictive control technology. Control Eng. Pract. 2003, 11,
733–764. [CrossRef]
25. Ochoa, S.G.; Wozny, G.; Repke, J.U. Plantwide Optimizing Control of a continuous bioethanol production
process. J. Process Control 2010, 20, 983–998. [CrossRef]
26. Idris, E.; Engell, S. Economics-based NMPC strategies for the operation and control of a continuous catalytic
distillation process. J. Process Control 2012, 22, 1832–1843. [CrossRef]
27. Takács, I.; Patry, G.G.; Nolasco, D. A dynamic model of the clarification thickening process. Water Res. 1991,
25, 1263–1271. [CrossRef]
28. Vilanova, R.; Santín, I.; Pedret, C. Control y Operación de Estaciones Depuradoras de Aguas Residuales:
Modelado y Simulación. Revista Iberoamericana de Automática e Informática Industrial RIAI 2017, 14, 217–233.
[CrossRef]
29. Cadet, C.; Bèteau, S.; Hernández, C. Multicriteria control strategy for cost/quality compromise in Wastewater
Treatment Plants. Control Eng. Pract. 2004, 12, 335–347. [CrossRef]
© 2017 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
3. Methodology Ti = qin xiin − qout xi (12)

x2 x9
Clearly, the problem stated previously falls within the realm 1 = H x6 (13)
of optimisation. Due to the discontinuous, semi-batch nature of KS + x2 KO + x9
the ATAD reaction, the process is time dependent. Thus, the above x2 x9
2 = H,T x7 (14)
problem has to be approached via dynamic optimisation. KS,T + x2 KO,T + x9
Generally speaking, there are two methods for dynamic opti-
misation: indirect and direct methods. For a recent review on 3 = bH x6 (15)
these two methods, see Bonvin, Palanki, and Srinivasan (2003). The
4 = bH,T x7 (16)
indirect method is perhaps less commonly used and involves an
analytical transformation and reformulation of the original prob- x4 /x6 x9
5 = kH x6 (17)
lem. Its general mathematical theory was developed by Pontryagin KX + x4 /x6 KO + x9
(1964). The more common direct methods, first used for aerospace
applications in the 1950s, involve a direct attack of the original x4 /x7 x9
6 = kH,T x7 (18)
problem. Direct methods can be subdivided into sequential and KX,T + x4 /x7 KO + x9
simultaneous approaches. The former, also called control vector x9
parameterisation (CVP), involves the parameterisation of the con- 7 = SP x7 (19)
KO,T + x9
trol variables (Banga, Balsa-Canto, & Moles, 2003); the latter, known
also as orthogonal collocation or direct transcription, involves where x1 represents the dissolved inert organic matter (gCOD/l),
parameterisation of both state and control variables (Biegler, 2007, x2 the readily biodegradable substrate (gCOD/l), x3 the particulate
2010). In this paper, we approach the above problem using CVP. inert organic matter (gCOD/l), x4 the slowly biodegradable sub-
To implement the optimisation problem at hand, we need a strate (gCOD/l), x5 the inactive thermophilic biomass (gCOD/l), x6
dynamic model of the ATAD reaction. In the subsequent sections, the active mesophilic biomass (gCOD/l), x7 the active thermophilic
we will introduce the model that was built for this purpose, fol- biomass (gCOD/l), x8 the inert organic matter from decay (gCOD/l),
lowed by a model assessment and parameter estimation, and a x9 dissolved oxygen (mgO2 /l), x10 the total enthalpy of the liquid
sensitivity analysis to identify the operating conditions with the phase (MJ), and x11 the total volume of the liquid phase (l). i are
strongest influence on the energy requirement. Finally, the prob- the conversion rates (gCOD/l day), while Ti represents the trans-
lem will be formulated following the CVP approach and the results port term for component i. qin and qout are the inlet and outlet
discussed for two different case studies. sludge flowrate (l/day), respectively. xiin represents the concen-
tration of component i in the inlet sludge. Values and formulae
3.1. ATAD model of the stoichiometric, kinetic, and thermodynamic parameters can
be found in Table 1. Values pertaining to reactor-specific param-
Generally speaking, in wastewater treatment, the systems of eters appearing in Eq. (10) are given later, depending on the case
ordinary differential equations describing mass and energy con- study. qa is the aeration flowrate (volumes of air per volumes of
versions are written in a compact form known as Petersen matrix. sludge per hour or vvh). Ka (volumes of sludge per volumes of air
However, we have decided to write the equations in the more con- or vv−1 ) is a parameter that has to be estimated for any particular
ventional form in order to conform to other papers dealing with plant and is, therefore, case-specific (Olsson & Newell, 1999). x9sat is
dynamic optimisation problems (e.g., Bonvin et al., 2003; Banga, the saturated dissolved oxygen concentration (mgO2 /l). For a more
Balsa-Canto, Moles, & Alonso, 2005). Also the notation has been detailed explanation of the model, see Rojas, Zhelev, and Bojarski
changed, but it is explained below. (2010). The presented model is based on the work by Gomez, de-
Gracia, Ayesa, and Garcia-Heras (2007) and Kovács, Csikor, Házi,
dx1
= T1 (1) and Miháltz (2007) and Kovács, Miháltz, and Csikor (2007), which
dt
in turn are extensions of the ASM1 model (Henze, Gujer, Mino, &
dx2 1 1 van Loosdrecht, 2000) at thermophilic temperatures.
= T2 − 1 − 2 + 5 + 6 (2)
dt YH YH,T
dx3 3.2. Treatment objectives
= T3 (3)
dt
dx4 The stabilisation degree or VS reduction rVS at a given time t is
= T4 + (1 − fP ) · 3 + (1 − fP,T ) · 3 − 5 − 6 (4) defined as
dt
dx5 XVS (t)
= T5 − 7 (5) rVS (t) = 1 − (20)
feed
dt XVS
dx6
= T6 + 1 − 3 (6) where XVS (t) is the VS concentration (g/l) within the reactor at time
dt feed
t (days), and XVS is the influent VS concentration (g/l). XVS can be
dx7 determined as the sum of all particulate components (Henze et al.,
= T7 + 2 − 4 + 7 (7)
dt 2000). The necessary condition to achieve a stable sludge is given
dx8 by
= T8 + fP · 3 + fP,T · 4 (8)
dt min
rVS − rVS ≤ 0 (21)
dx9 1 − YH 1 − YH,T
= T9 − 1 − 2 + Ka qa (x9sat − x9 ) (9)
dt YH YH,T Countries with different regulations may require different stabilisa-
tion levels to be achieved before discharge. This level also depends
dx10 1 − YH
= T10 + Yheat 1 x11 + P − Pwall + Qs + Ql (10) on the desired quality of the end product. In the US and Europe,
dt YH
the required level for Class A Biosolids is 38% (USEPA, 1993). That
dx11 min = 0.38. The minimum time required to satisfy Eq. (21) is
is, rVS
= T11 (11)
dt defined here as the stabilisation time tS (days).
Table 1
Values and formulae of stoichiometric, kinetic, and thermodynamic parameters used in the ATAD model (see Eqs. (1)–(19)).
Parameter Value/formula Note Source
Stoichiometrica
YH (gCOD/gCOD) 0.85 Mesophilic growth yield Kovács, Miháltz, et al. (2007)
YH,T (gCOD/gCOD) 0.8 Thermophilic growth yield Kovács, Miháltz, et al. (2007)
fP (gCOD/gCOD) 0.0092 · T − 0.16 Fraction of inert COD generated by mesophilic lysis Kovács, Miháltz, et al. (2007)
fP,T (gCOD/gCOD) 0.2 Fraction of inert COD generated by thermophilic lysis Kovács, Miháltz, et al. (2007)
Kinetica
H (day−1 ) 72 · e(−42.95/T) − 201.5 · e(−135.6/T) Mesophilic growth rate Kovács, Miháltz, et al. (2007)
H,T (day−1 ) 155.4 · e(−6.086/(T−34)) − 132.8 · e(−7.734/(T−34)) Thermophilic growth rate Kovács, Miháltz, et al. (2007)
bH (day−1 ) 1075 · e(−159.9/T) Mesophilic decay rate Kovács, Miháltz, et al. (2007)
bH,T (day−1 ) 0.001 · T2 + 0.11 · T − 2.8 Thermophilic decay rate Kovács, Miháltz, et al. (2007)
KS (g/l) 4.5 · T − 80 Mesophilic substrate half-saturation constant Kovács, Miháltz, et al. (2007)
KS,T (g/l) e(0.13·T−2.5) Thermophilic substrate half-saturation constant Kovács, Miháltz, et al. (2007)
KO (g/l) 0.17 Mesophilic oxygen half-saturation constant Kovács, Miháltz, et al. (2007)
KO,T (g/l) 0.17 Thermophilic oxygen half-saturation constant Kovács, Miháltz, et al. (2007)
KX (gCOD/gCOD) e(0.255·T−9.9) Mesophilic hydrolysis half-saturation constant Kovács, Miháltz, et al. (2007)
KX,T (gCOD/gCOD) 6.4 Thermophilic hydrolysis half-saturation constant Kovács, Miháltz, et al. (2007)
kH (day−1 ) 42 · e(−45.57/T) − 12 · e(−363.8/T) Mesophilic maximum specific hydrolysis rate Kovács, Miháltz, et al. (2007)
kH,T (day−1 ) 6.872 + e(0.196·T−8.956) Thermophilic maximum specific hydrolysis rate Kovács, Miháltz, et al. (2007)
SP (day−1 ) max
SP
/(1 + (max
SP
/0SP − 1) · e(−max
SP
· ı · )) Specific constant rate of thermophilic activation Kovács, Csikor, et al. (2007)
max
SP
(day−1 ) −0.282 · T2 + 28.7 · T − 635 Maximum activation rate Kovács, Csikor, et al. (2007)
0SP (day−1 ) −0.000154 · T + 0.016107 · T − 0.363
2
Minimal activation rate Kovács, Csikor, et al. (2007)
ı 0.5 Rapidity of change of SP Kovács, Csikor, et al. (2007)
Thermodynamic
YHeat (MJ/kgO2 ) 13.9 Specific heat yield of thermophiles Gomez et al. (2007)
x9sat (mg/l) 10 Dissolved oxygen saturation concentration By default
f (g COD/g VS) 1.42 Conversion factor between COD and VS USEPA (1990)
a
Formulae of stoichiometric and kinetic parameters were modified in order to allow a better fit with the data presented by Kovács, Miháltz, et al. (2007). Thus, the formulae
presented herein and those presented by Kovács, Miháltz, et al. (2007) may differ. For temperature-dependent parameters, the temperature is given in ◦ C, and they are only
valid for temperatures higher than 35 ◦ C.
tr
On the other hand, the degree of pasteurisation or pasteurisation 1 P=const P· tr
EV = P(t) · dt = , (28)
lethality LP at time t is defined by Eq. (22), the so-called general Vin Vin
0
method (Toledo, 1991).
Vin
t CV = , (29)
d tr
LP (t, t0 ) = (22)
t0
D where min represents the total mass of VS in the influent sludge (kg),
P the power of the aeration equipment (kW), and Vin the volume of
influent sludge (m3 ). Em and EV represent the gravimetric (kWh per
50 070 000
D= for T ≥ 50 ◦ C and D ≥ 0.021 days (30 min) kg of VS treated) and volumetric (kWh per m3 of sludge treated)
100.14T energy requirements, respectively. Cm and CV are the gravimetric
(23)
(kg of VS treated per day) and volumetric (m3 of sludge treated per
Eq. (23) expresses the relationship between temperature T and day) plant capacity, respectively.
its corresponding contact time or thermal death time D (days) We do not consider the volumetric magnitudes to be adequate
required to achieve pasteurisation (USEPA, 1993). Note that, while measures or indicators, as they do not take into consideration the
the degree of stabilisation rVS is an instantaneous measure, the influent concentration of VS, which are in fact the subject of the
degree of pasteurisation LP depends on the temperature history treatment. Thus, it should be understood that in the rest of this
within the interval [t0 , t]. The necessary condition to achieve a paper when speaking about energy requirement or plant capacity
pasteurised sludge is expressed by Eq. (24). we refer to the gravimetric magnitudes, unless otherwise speci-
fied. Note that, being P the power of the aeration equipment, it
1 − LP ≤ 0 (24) is implied that we are exclusively considering the energy require-
ments in terms of aeration.
The minimum time required to satisfy this condition is defined
here as the pasteurisation time tP (days). The overall reaction time 4. Case studies
tr (days) is then determined by the greatest of these two times:
tr = max{ts , tp }. (25) Two designs will be considered for optimisation: a single-stage

system (case study 1), and a two-stage system (case study 2). In
both cases, ATAD works as single sludge treatment (i.e., no Dual sys-
3.3. Energy requirement and plant capacity tem), and the sludge source are the primary and secondary stages
of wastewater treatment. The most important design and operat-
The specific energy requirement and plant capacity of the treat- ing parameters of the two case study plants are shown in Table 2.
ment can be determined as A brief description of the two designs follows:
tr
1 P=const P · tr
Em = P(t) · dt = , (26) • Case study 1 (CS1). This single-stage system consists of a 300-
min 0
min
m3 cylindrical reactor (Fig. 1a). With a typical sludge flowrate of
min 30 m3 /day, a batch time of 1 day, and an influent VS concentration
Cm = , (27)
tr of 34 g/l, the energy requirement is 0.72 kWh/kg. The aeration
Fig. 1. Photographs of (a) single-stage system CS1 (Gomez, 2007), and (b) two-stage system CS2 (Rojas, 2011).
flowrate is relatively low with 0.8–1.4 vvh using a Venturi sys- concentrations was used. The data set was taken from Gomez
tem. The plant is located in Navarra (Spain) and it is described in (2007) and Gomez et al. (2007). Fig. 2 shows the MAE for prediction
more detail in Gomez (2007) and Gomez et al. (2007). The relation of temperature MAET and VS concentration MAEVS , as well as the
between power and aeration flowrate is given by P = 28 + 6 · qa simulation of temperature and VS concentration for the optimal
where P is in kW and qa in vvh. Ka -value of 5 vv−1 . Optimal Ka -values for MAET (5.3 vv−1 ) and
• Case study 2 (CS2). This system consists of two cylindrical 100- MAEVS (4.9 vv−1 ) differ slightly. Thus, the value of 5 vv−1 was taken
m3 reactors in series. An important feature of this plant is that as a compromise. Both MAET and MAEVS are highly sensitive to
it makes use of a polymer as pre-thickening agent. The reason changes of Ka . Data simulation with the optimal Ka -value in Fig. 2c
for this is that the influent sludge has a total VS concentra- and d shows the fit is qualitatively reasonable even though there
tion of just 15 g/l which is not sufficiently high to reach and is still room for improvement, in quantitative terms. We see, at
maintain thermophilic temperatures. Thus, polymer is added to least, three potential sources for the deviations observed in Fig. 2c
reach a concentration of 40 g/l. With a typical sludge flowrate and d between experimental data and model predictions:
of 25 m3 /day, a batch time of 1 day, and an influent VS concen-
tration of 15 g/l (as polymer is not the object of the treatment • The assumption of linear dependence of the aeration term g =
but only the natural solids), the minimum energy requirement is Ka qa (x9sat − x9 ) on the reactor-specific parameter Ka and the aera-
1.22 kWh/day. The aeration flowrate is constant at 4 vvh (11 kW) tion flowrate qa . As pointed out by Olsson and Newell (1999), this
in the first stage reactor and 3 vvh (9 kW) in the second. A linear linear dependence is just the simplest one, while other more com-
extrapolation of these values (i.e., P1 = 11 · qa1 /4 and P2 = 9 · qa2 /3) plex relationships are possible. Future model validations should
was assumed when using other aeration levels. This two-stage therefore include a thorough investigation of the effect of the dif-
design is perhaps the most widely used and it is described in ferent kinds of dependence g(Ka , qa ) on the accuracy of model
detail in USEPA (1990). The plant, located in Killarney (Ireland), predictions.
is shown in Fig. 1b. • The limited accuracy of the estimation of thermophilic kinetic and
stoichiometric parameters by Kovács, Miháltz, et al. (2007), con-
5. Model assessment and parameter estimation sidering that they significantly diluted the sludge samples used
for calibration in order to carry out the necessary respiromet-
The objective of this section is to estimate the value of the case- ric measurements. Under such conditions it is unclear whether
specific parameter Ka for the two different ATAD designs, and to the obtained parameter values have not been used here beyond
assess the adequacy of the ATAD model. The value of Ka was deter- their domain of validity. It should be highlighted that the afore-
mined through the minimisation of the mean absolute error (MAE). mentioned study is the only one known to the authors that
In the case of the single-stage system CS1, a 21-day time estimates the value of stoichiometric and kinetic parameters
series comprising measurements of both temperature and VS of thermophilic microorganisms. Thus, more work is needed to
further and more accurately estimate thermophilic parameters,
Table 2 especially under conditions that are closer to those actually found
Design and operating parameters of the three case study plants. in ATAD systems.
Parameter Value Value
• A part of the experimental input data (corresponding to week-
CS1a CS2b ends) was not measured and it had thus to be estimated. This
No. of reactors in series 1 2
could be responsible for, at least, part of the deviations observed
Reactor volume (m3 ) 300 100 in Fig. 2c and d.
Gravimetric energy requirement (kWh/kg VS) 0.72 1.22
Daily load (m3 /day) 30 25
In the case of the two-stage system CS2, a 5-day temperature
Batch time (h) 24 24
Reaction time (h) 22.8 23
time series was used to estimate the value of Ka . No regular mea-
Hydraulic retention time (days) 5 7–10 surements of the VS concentration were available. Fig. 3a shows the
Specific power (W/m3 ) 119 100 MAE for prediction of temperature. The function reaches a min-
Aeration flowrate (vvh) 0.8–1.4 3–4 imum at 3.5 vv−1 . Compared to Fig. 2a, it can be seen that the
Influent VS concentration before pre-thickening (g/l) – 15
minimum MAE value is relatively lower, but the MAE is still highly
Influent VS concentration before treatment (g/l) 36 40
Influent temperature (◦ C) 10 10 sensitive to changes of Ka . The higher accuracy in this case may be
Temperature in first reactor (◦ C) 50–65 40–50 due to smaller changes in the operating conditions. Fig. 3b shows
Temperature in second reactor (◦ C) – 50–65 the simulated temperature time series for the optimal Ka -value of
a
Adapted from Gomez et al. (2007) and Gomez (2007). 3.5 vv−1 . The fit is reasonably good both in qualitative and quanti-
b
Adapted from Rojas (2011). tative terms.
(a) (b) 2.2

7
2
6
MAE VS (g/l)
MAE T (ºC)
1.8
5
1.6
4 1.4
1.2
3
3 4 5 6 7 3 4 5 6 7
Ka (vv-1) Ka (vv-1)
(c) 75 (d) 20
Data
VS concentration (g/l)
18
Temperature (ºC)
70 Simulation
16
65
14
60
12
55 Data
10
Simulation
50 8
0 5 10 15 20 5 10 15 20
Fig. 2. Model assessment and estimation of Ka -value for CS1: MAE for prediction of (a) temperature MAET and (b) VS concentration MAEVS as a function of Ka , and simulation
of (c) temperature and (d) VS concentration for optimal Ka -value (5 vv−1 ).
(a) 7 (b) 70
65
6
60
Temperature (ºC)
MAET (oC)
5 55
50
4
45
3 40
35
0 1 2 3 4 5
2
2 3 4 5 6 Time (days)
Ka (vv-1) DataR1 DataR2 SimR1 SimR2
Fig. 3. Model assessment and estimation of Ka -value for CS2: (a) MAE for prediction of temperature time series as a function of Ka , and (b) simulation of temperature time
series for optimal Ka -value (3.5 vv−1 ).
6. Sensitivity analysis same set of operating conditions. Reaction time, energy require-
ment, and plant capacity are then determined based on the last
A global sensitivity analysis was performed to examine the effect batch of the series.
of the operating conditions on the energy requirement and thus to Fig. 4 shows the effect of the considered operating conditions
identify the most promising optimisation variables. on the energy requirement. Some of the correlations have been
The modelled system consists of a single 100-m3 reactor with attached a linear fit. It should be clarified that this is not to
the characteristics described in USEPA (1990). The considered oper- imply that a first-order correlation underlies such disparate data,
ating conditions include the aeration flowrate, sludge flowrate but rather to show, on a merely qualitative basis, that a posi-
(through the loading time and the percent of volume replaced), and tive (increasing) or negative (decreasing) correlation underlies the
influent temperature. For a given set of operating conditions, the information in question. In this way, the linear fit should only be
model determines the reaction time, and then calculates the energy seen as a qualitative aid for what might otherwise be seen with the
requirement and plant capacity. The reaction time is thus implicitly naked eye.
considered as an operating condition even though it cannot be set Fig. 4a displays the effect of the aeration flowrate, which
beforehand. As a means to ensure periodic solutions (for reasons correlates positively with Em. This is highlighted by red line which
explained later), nb consecutive batches are calculated using the corresponds to the linear fit between the two variables. This
(a) x2 x8 x4 x5 x7 (b)

2 0.5
Concentrations
0.4
1.5
0.3
1
0.2
0.5
0.1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
(c) 72 (d) 12 400

350
Temperature (ºC)
10
OUR (mgO2/l h)
70 300
8 250
68
200
6
66 150
4 100
64 2 50
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Fig. 5. Optimal solution for CS1: (a) non-dimensional concentrations, (b) dissolved oxygen concentration, (c) temperature, and (d) aeration flowrate and OUR (N = 20). The
optimal energy requirement is 0.56 kWh/kg (22% improvement).
Let us consider a series of nb consecutive batches for a given Fig. 5 displays the optimal solution found for CS1 with an energy
control vector u, a given initial condition of the first batch x10 , and requirement of 0.56 kWh/kg which represents a 22% improvement.
fixed load and characteristics of the influent sludge (i.e., no dis- As it can be seen from the non-dimensional concentrations in Fig. 5a
turbances). Generally speaking, in this scenario the state variables (see initial condition and input in Appendix A), hydrolysis and
will tend to be non-periodic for the first batches in the series and stabilisation are the rate-limiting factors of the reaction, that is,
quasi-periodic for the later batches (at least for sufficiently large rVS (tf ) = 0.38. This was the case for almost all optimal solutions.
nb -values). Consequently, it can happen that some of the path con- This tendency is consistent with our previous finding that the sta-
straints are satisfied for the first batches in the series but not for bilisation process is generally the limiting factor of the reaction
the last ones, or vice versa. Thus, if we evaluate the path constraints (Rojas et al., 2010). Fig. 5b and c shows the optimal DO and tem-
based on one single batch, chances are high that we obtain some perature profiles. DO is consistently below 1 mg/l despite the high
solutions that are not feasible. To surmount this difficulty, for a aeration levels, meaning that the OURs are very high. Operating
given iteration within the optimisation loop, the code evaluates temperatures are high enough to exceed pasteurisation require-
the path constraints based on the last of nb consecutive batches ments (LP (tf ) 1). This was found to be the case in most optimal
using for all of them the same control vector u. This procedure will solutions. Fig. 5d displays the optimal profiles of the aeration
ensure the periodicity of the solutions and prevent us from obtain- flowrate (N = 20) and OUR. It can be seen that the values of the
ing unfeasible solutions. According to our experience, it is sufficient aeration are generally much higher than those used in the plant
to set nb equal to or higher than 10 to ensure periodicity. Thus, in (up to 12 vvh). Even though there is an apparent randomness in
the following implementations the value 10 was chosen. its behaviour, the general tendency of the aeration flowrate is to
It should also be noted that for simplicity the reactor loading was start from higher values and to gradually decrease until reach-
regarded as instantaneous. This implies a constant volume of sludge, ing the lower boundary at the end of the reaction. It is unclear,
and allows us to implement the optimisation in a computationally though, what may have caused the algorithm to choose this rela-
much more efficient manner. tively erratic and “spiky” path. However, the qualitative behaviour
of the aeration flowrate is biologically plausible: at the beginning
7.2.1. Optimisation results CS1 of the reaction (i.e. right after reactor loading) the concentration
The final time (days) was bound within the interval [0.7, 1.5], of readily biodegradable organic matter x2 is highest; this drives
and the aeration flowrate (vvh) within [0.8, 15]. The reason for the the system to a high demand of oxygen needed to oxidise the sub-
lower boundary of the final time is that the actual plant has a max- strate; in order to match the high oxygen demand, the aeration
imum sludge production of ∼42 m3 /day; with a 10% displacement flowrate has to be higher at the beginning of the batch. As time
of the reactor content after each batch (i.e., 30 m3 ), this would allow passes and substrate is gradually consumed, the demand for oxy-
a minimum reaction time of 0.7 days. The boundaries of the aer- gen decreases and so does the aeration flowrate. The behaviour
ation flowrate have been chosen so as to allow wide variations, of the OUR as well as that of DO evidently represent a reaction
much higher than those seen in the actual plant (i.e., [0.8, 1.4]). to changes in the aeration flowrate. Thus, the qualitative optimal
The lower boundary was kept at 0.8 vvh, as a minimum, nonzero profile of the aeration flowrate makes sense of the underlying bio-
level of aeration is required in this type of equipment. Note that logical system. This result also shows that the optimal behaviour of
the power-aeration relation is given by P = 28 + 6 · qa , where P is the the aeration flowrate is far from invariable, as it is typically found in
power (kW) at the aeration level qa (vvh). This relation holds in the conventional ATAD systems. At this stage, it can thus be seen that
aeration interval [0.8, 1.4] and it was assumed that it also holds in exploiting the aeration flowrate as an additional degree of free-
the interval [0.8, 15]. dom for optimisation is a crucial way of obtaining better solutions,
(b) 1
x2 x8 x4 x5 x7
R1
(a) R2

0.8
1.5
Concentrations
0.6
1
0.4
0.5
0.2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

(c) 60 (d) 10 500

55 8 400
Temperature (ºC)
R1
h)
R2
OUR (mgO/l
2
50 6 300
45 4 200
40 2 100
35
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Fig. 6. Optimal solution for CS2: (a) non-dimensional concentrations (solid lines: Reactor 1; dashed lines: Reactor 2), (b) dissolved oxygen concentration, (c) temperature,
and (d) aeration flowrate and OUR (N1 = 10, N2 = 10) (solid lines: Reactor 1; dashed lines: Reactor 2). The optimal energy requirement is 0.99 kWh/kg (18% improvement).
and that invariable aeration is not generally suitable for ATAD sys- as optimisation variable within the interval [0, 30]. Note that this
tems. polymer is added to the influent sludge which has an original con-
The optimal reaction time tf was found at the lower boundary centration of 15 g/l. Thus, after pre-thickening the minimum and
of 0.7 days (the typical time of operation being 1 day). It was found maximum VS concentrations are 15 and 45 g/l, respectively.
that the algorithm tends to favour shorter final times, which is also The optimal solution found for CS2 is displayed in Fig. 6 with an
consistent with the aforementioned sensitivity analysis. Thus, the energy requirement of 0.99 kWh/kg which represents an improve-
optimal value of the final time, being the value of the lower bound- ment of 18%. The concentrations in Fig. 6a (see initial condition and
ary, makes sense of the dependence of the energy requirement input in Appendix A) show that hydrolysis and stabilisation are,
(objective functional) on the final time: shorter final times tend to once more, limiting. As in the previous case study, DO in Fig. 6b is in
result in lower energy requirements. Let us recall that conventional both reactors consistently below 1 mg/l. Even though the temper-
systems use a 1-day final time due to operational convenience, thus ature of the first-stage reactor is relatively low, the second-stage
not exploiting this important degree of freedom. Even though in reactor operates consistently at thermophilic temperatures (see
some optimal solutions tf was between the boundaries, in most (and Fig. 5c), and pasteurisation requirements are exceeded. The overall
actually the best) of them it was found at the lower boundary. This low temperatures are a result of the low optimal polymer concen-
seems to suggest that, when minimising the energy requirements, tration (6.5 g/l).
lower reaction times and higher aeration flowrates are favoured. The optimal aeration profiles (N1 = 10, N2 = 10) in Fig. 6d show
It should also be noted that the objective function is multimodal that, with exception of “spiky” intervals, the aeration levels are
as several, diverse solutions were found. Thus, by no means we can generally lower than in the current operation of the plant. It does
guarantee the global optimality of the solution. not seem, in principle, straightforward what might be the biolog-
ical plausibility of the optimal aeration profile for the first-stage
7.2.2. Optimisation results CS2 reactor, which remains nearly constant throughout the batch with
The search interval for the final time was [0.5, 1.5], and for the the exception of the spikes. However, in the case of the second
aeration flowrate [0.8, 15]. Again, the boundaries of the aeration stage (and excluding the spikes), the aeration flowrate is higher
flowrate have been chosen so as to allow wide variations, much at the beginning and lower at the end of the batch. This is the
higher than those seen in the actual plant (i.e., [3, 4]). The lower same qualitative solution that was found for CS1. Nevertheless, it is
boundary was also kept at 0.8 vvh, as aeration significantly con- not clear that the same biological plausibility applies here, because
tributes to sludge mixing in this kind of design. In this design, the second-stage reactor has a very low concentration of substrate
the first-stage reactor has an aeration power of 11 kW at a con- throughout the reaction. Again, it is unclear what may have lead the
stant aeration level of 4 vvh and the second-stage reactor 9 kW at algorithm to chose values pertaining to the “spiky” regions of the
3 vvh. Thus, the power-aeration relation over the aeration interval aeration flowrate. Generally, the algorithm did not obtain smooth
[0.8, 15] was assumed to be P1 = (11/4) · qa,1 for the first reactor and trajectories for the aeration flowrate. It can be observed that in
P2 = (9/3) · qa,2 for the second, where P is the power (kW) at the aer- this case lower aeration flowrates, lower polymer concentrations,
ation level qa (vvh). The polymer concentration (g/l) was also used and higher reaction times were favoured to minimise the energy
Table 3 reactors was found to be lower than currently used. This means
Values of energy requirement, final time, and polymer concentration, prior and after
that it would be enough to keep it at about 2–3 vvh for one day to
optimisation for the two case-study plants.
satisfy treatment objectives.
Parameter CS1 CS2 Current ATAD models, including the one presented here, are
Base energy value (kWh/kg) 0.72 1.22 still in a relatively early stage of development when compared to
Minimum energy value (kWh/kg) 0.56 0.99 ASM models. To improve the accuracy of these models and that
Energy saving (%) 22 18 of optimisation results, further assessment, parameter estimation,
Base final time (days) 1 1
and model validation are needed. For this purpose, more quantity
Optimal final time (days) 0.7 1.08
Base polymer concentration (g/l) – 25 and higher quality full-scale data, at present extremely scarce, have
Opt. polymer concentration (g/l) – 6.5 to be made available.
Despite its advantages and common use in other fields, opti-
misation is still relatively rare in wastewater engineering. Given
requirement. It is also remarkable that the OUR of the first-stage the high, rising cost of wastewater treatment, optimisation should
reactor presents in this case a certain delay with respect to changes become the norm when it comes to design and operation of
in the aeration flowrate. wastewater treatment plants.
The optimal polymer concentration is 6.5 g/l and the total VS
concentration in the influent sludge, 21.5 g/l. This value goes is
lower than the minimum 25 g/l recommended by USEPA (1990) Acknowledgements
to consistently operate at thermophilic temperatures. However, as
shown by Fig. 6c the second-stage reactor consistently operates We wish to thank Jairo Gomez (NILSA, Navarra, Spain) for pro-
in the thermophilic temperature range. This result shows that the viding data and for his continued support. This publication has
assumption that VS concentration must be over 25 g/l to reach and emanated from research conducted with the financial support of
maintain thermophilic operation does not generally hold. Science Foundation Ireland under Grant No. 06/CP/E007.
The reaction time was slightly higher (1.08 days) than in the
typical 1-day operation. We noticed that this was the case for
Appendix A.
all optimal solutions in this particular case study. The algorithm
tended to find solutions close to this value, even when the start-
A.1. Details on optimisation results of CS1 shown in Fig. 5
ing point was far away from it or when starting with a randomized
matrix of starting points. This is consistent with the suggestion of
The initial condition of the system corresponding to the solution
Scisson (2003) that some ATAD systems may require longer (than 1
displayed in Fig. 5 is as follows:
day) final times to guarantee compliance with stabilisation require-
ments. It is not clear why in this particular case the algorithm did
x0 = [3, 1.5, 3, 7.8, 0.47, 0, 0.83, 8, 0, 62.8, 300]
not seek to further shorten the final time.
Table 3 summarises base and minimum energy values, the base
Note that all concentrations (i.e. from x1 to x9 ) are in g/l. x10
and optimal final time, and the energy savings for each of the two
stands here for the temperature (◦ C) instead of the total enthalpy
case studies.
of the liquid phase, and x11 for the total volume (m3 ).
As for the inputs, the amount of sludge replaced after each batch
8. Conclusions was 30 m3 (10% of total working volume). The total VS concentra-
tion in the influent sludge was 36.9 g/l, that is, the same as in the
The aim of this paper was to minimise the energy requirement of actual plant. The distribution of mass among the different compo-
ATAD while complying with treatment objectives, i.e. sludge stabil- nents in the influent sludge was as follows: x1 = 2.7 g/l, x2 = 4.5 g/l,
isation and pasteurisation. To this purpose, a dynamic ATAD model x3 = 2.7 g/l, x4 = 22.5 g/l, x5 = 3.1 g/l, x6 = 0.1 g/l, x7 = 1.35 g/l, x8 = 0 g/l,
was presented and assessed. A global sensitivity analysis was per- x9 = 0.6 mg/l, x10 = 10 ◦ C. The aeration flowrate is displayed in
formed to identify promising optimisation variables, that is, those Fig. 5d.
with the strongest influence on the energy requirement. It was
found that the aeration flowrate, the reaction time, and the sludge
flowrate (percentage of reactor volume replaced, loading time, and A.2. Details on optimisation results of CS2 shown in Fig. 6
reaction time) had a clear impact on the energy requirement, while
the influent temperature did not. As for the optimisation problem, The initial condition of the system corresponding to the solution
it was formulated following the sequential approach for dynamic displayed in Fig. 6 is as follows:
optimisation, due to the discontinuous, semi-batch nature of ATAD.
Two full-scale ATAD facilities were selected as case studies: a one- x01 = [0, 4.2, 0, 10.9, 0.01, 0, 1.82, 2.1, 0, 32.7, 100]
stage and a two-stage system. For the single-stage system, a 22%
improvement of the energy requirement was reached after opti-
misation, when compared to current plant operation. Similarly, for x02 = [0, 0.75, 0, 7.3, 0, 0, 1, 5.7, 0, 54.9, 100]
the two-stage system an improvement of 18% was attained. The
objective function was found to be highly multimodal in both case All concentrations (i.e. from x1 to x9 ) are in g/l. x10 stands here
studies, and the global optimality of the solutions can thus not be for the temperature (◦ C), and x11 for the total volume (m3 ).
guaranteed. As for the inputs, the amount of sludge replaced after each batch
Based on the optimisation results, the following recommen- was 25 m3 (12.5% of total working volume). The total VS concen-
dations can be made. For the single-stage system, the qualitative tration in the influent sludge was 21.5 g/l, that is, about the half
solution is obtained by applying a higher aeration flowrate at the as in the actual plant. The distribution of mass among the differ-
beginning of the reaction and decreasing it gradually until reaching ent components in the influent sludge was as follows: x1 = 0 g/l,
the lower boundary at the end of the batch. The final time should x2 = 9.2 g/l, x3 = 0 g/l, x4 = 11.7 g/l, x5 = 0 g/l, x6 = 0.1 g/l, x7 = 0.5 g/l,
be kept at its lowest value, corresponding to the highest sludge x8 = 0 g/l, x9 = 0.6 mg/l, x10 = 10 ◦ C. The aeration flowrate is displayed
flowrate. For the two-stage system, the aeration level in both in Fig. 6d.
References Kelly, H. G., & Mavinic, D. S. (2003). Autothermal thermophilic aerobic digestion
research, application and operational experience. In WEFTEC 2003 workshop
Banga, J. R., Balsa-Canto, E., & Moles, C. G. (2003). Dynamic optimization of biore- W104 Los Angeles, CA, October 2003.
actors: A review. Proceedings of the Indian National Science Academy, 69A, Kovács, R., Csikor, Zs., Házi, F., & Miháltz, P. (2007). Application of activated sludge
257–265. model no. 3 for the modeling of organic matter biodegradation at thermophilic
Banga, J. R., Balsa-Canto, E., Moles, C. G., & Alonso, A. A. (2005). Dynamic opti- temperatures. Water Environment Research, 79(5), 554–560.
mization of bioprocesses: Efficient and robust numerical strategies. Journal of Kovács, R., Miháltz, P., & Csikor, Zs. (2007). Kinetics of autothermal thermophilic
Biotechnology, 117(4), 407–419. aerobic digestion: Application and extension of activated sludge model no. 1 at
Barnes, D., Bliss, P. J., Gould, B. W., & Vallentine, H. R. (1986). Water and wastewater thermophilic temperatures. Water Science & Technology, 56(9), 137–145.
engineering systems. Longman Scientific & Technical. LaPara, T. M., & Alleman, J. E. (1999). Thermophilic aerobic biological wastewater
Batts, C. W., Burton, F. L., & Jones, M. (1993). Demand side management opportunities treatment. Water Research, 33(4), 895–908.
in the wastewater industry. In Water environment federation, proceedings of the Layden, N. M., Kelly, H. G., Mavinic, D. S., Moles, R., & Barlet, J. (2007a). Autother-
66th annual conference & exposition. mal thermophilic aerobic digestion (ATAD) – Part I: Review of origins, design,
Biegler, L. T. (2007). An overview of simultaneous strategies for dynamic optimiza- and process operation. Journal of Environmental Engineering and Science, 6(6),
tion. Chemical Engineering and Processing, 46, 1043–1053. 665–678.
Biegler, L. T. (2010). Nonlinear programming: Concepts algorithms and applications to Layden, N. M., Kelly, H. G., Mavinic, D. S., Moles, R., & Barlet, J. (2007b). Autothermal
chemical processes. SIAM—Society for Industrial and Applied Mathematics. thermophilic aerobic digestion (ATAD) – Part II: Review of research and full-
Bonvin, D., Palanki, S., & Srinivasan, B. (2003). Dynamic optimization of batch pro- scale operating experiences. Journal of Environmental Engineering and Science,
cesses: I. Characterization of the nominal solution. Computers and Chemical 6(6), 679–690.
Engineering, 27, 1–26. Le, S. M. (2006). Thermophilic biological pre-treatments for MADs. In AquaEnviro
Burton, F. L. (1996). Water and wastewater industries: Characteristics and energy workshop: Advances in technology for the anaerobic digestion of municipal sludge
management opportunities, CR-10691. St. Louis, MO: Electric Power Research Manchester, UK, 14 June 2006.
Institute. M/J Industrial Solutions. (2003). Municipal wastewater treatment plant energy base-
Commission of the European Communities. (2006). Action plan for energy effi- line study. Prepared for Pacific Gas and Electric Company, San Francisco CA.
ciency: Realising the potential. Technical Report COM(2006)545 final. Brussels: Metcalf and Eddy, Inc. (2003). Wastewater engineering: Treatment and reuse (4th ed.).
European Commission. New York: McGraw-Hill.
Csikor, Z., Mihaltz, P., Hanifa, A., Kovacs, R., & Dahab, M. F. (2002). Identification of fac- Moles, C. G., Gutierrez, G., Alonso, A. A., & Banga, J. R. (2003). Integrated process
tors contributing to degradation in autothermal thermophilic sludge digestion. design and control via global optimization: A wastewater treatment plant case
Water Science and Technology, 46(10), 131–138. study. Chemical Engineering Research and Design, 81, 507–517.
Descoins, N., Stephane, D., Remi, L., & Marechal, F. (2010). Energetic efficiency in Olsson, G., & Newell, B. (1999). Wastewater treatment systems. Modelling, diagnosis
Waste Water Treatments Plants: Optimization of activated sludge process cou- and control. London, UK: IWA Publishing.
pled with anaerobic digestion. In Proc. of the 23rd international conference on Ponti, C., Sonnleitner, B., & Fiechter, A. (1995a). Aerobic thermophilic treatment of
efficiency, cost, optimization, simulation and environmental impact of energy sys- sewage sludge at pilot plant scale. 1: Operating conditions. Journal of Biotech-
tems Lausanne, 14–17 June, (pp. 1062–1069). nology, 38(2), 173–182.
Egea, J. A., Balsa-Canto, E., García, M. G., & Banga, J. R. (2009). Dynamic optimization Ponti, C., Sonnleitner, B., & Fiechter, A. (1995b). Aerobic thermophilic treatment of
of nonlinear processes with an enhanced scatter search method. Industrial and sewage sludge at pilot plant scale. 2: Technical solutions and process design.
Engineering Chemistry Research, 48(9), 4388–4401. Journal of Biotechnology, 38(2), 183–192.
Egea, J. A., Martí, R., & Banga, J. R. (2010). An evolutionary method for Pontryagin, L. S. (1964). Mathematische Theorie Optimaler Prozesse. Wien: Olden-
complex-process optimization? Computers and Operations Research, 37(2), bourg Verlag.
315–324. Rojas, J. (2011). Energy efficiency optimisation of wastewater treatment: Study of
Fikar, M., Chachuat, B., & Latifi, M. A. (2005). Optimal operation of alternating acti- ATAD. Ph.D. Thesis, University of Limerick, Limerick, Ireland.
vated sludge processes. Control Engineering Practice, 13(7), 853–861. Rojas, J., Zhelev, T., & Bojarski, A. D. (2010). Modelling and sensitivity analysis of
Focus on Energy. (2006). Water and wastewater energy best practice guidebook. Pre- ATAD. Computers and Chemical Engineering, 34(5), 802–811.
pared for Wisconsin Department of Administration by the Focus on Energy Scisson, J. P. (2003). ATAD, the next generation: Design, construction, start-up and
Program. Madison, WI, US. operation of the first municipal 2nd generation ATAD. In Joint residuals and
Gomez, J. M. (2007). Digestion Aerobica Termofila Autosostenida (atad) de fangos: biosolids management conference and exhibition 2003.
Estudio experimental a escala real y modelizacion matematica del reactor. Ph.D. Toledo, R. T. (1991). Fundamentals of food process engineering (2nd ed.). New York:
Thesis, Universidad de Navarra, San Sebastian, Spain. Chapman and Hall.
Gomez, J., de-Gracia, M., Ayesa, E., & Garcia-Heras, J. L. (2007). Mathematical mod- USEPA. (1990). Environmental regulations and technology: Autothermal ther-
elling of autothermal thermophilic aerobic digesters. Water Research, 41(5), mophilic aerobic digestion of municipal wastewater sludge. Technical report,
959–968. United States Environmental Protection Agency, Office of Research and Devel-
Henze, M., Gujer, W., Mino, T., & van Loosdrecht, M. C. M. (2000). Activated sludge opment, Washington, DC.
models ASM1, ASM2, ASM2d and ASM3. Scientific and Technical Report No. 9. USEPA. (1993). 40 CFR PART 503. Standards for the use or disposal of sewage
London: IWA Publishing. sludge. Washington, DC: United States Environmental Protection Agency, Office
Holenda, B., Domokos, E., Redey, A., & Fazakas, J. (2007). Aeration optimization of of Research and Development.
a wastewater treatment plant using genetic algorithm. Optimal Control Applica- Warakomski, A., McWhirter, J., Balan, P., Swope, B., & Gyger, R. (2007). Evolution of
tions and Methods, 28, 191–208. atad/dual digestion and recent process developments. In IWA specialist confer-
Jones, M. (1991). Demand side management: Opportunities in water and sewage ence: moving forward wastewater biosolids sustainability: Technical, managerial,
treatment. IEEE Power Engineering Review, 11(11), 8–9. and public synergy Moncton, New Brunswick, Canada, June 2007.
Juteau, P. (2006). Review of the use of aerobic thermophilic bioprocesses for the Yonkin, M., Clubine, K., & O’Connor, K. (2008). Importance of energy efficiency to the
treatment of swine waste. Livestock Science, 102(3), 187–196. water and wastewater sector. Clear Waters, 38, 12–13.

Nutrition Community Era College Wastewater

Uploaded by

Copyright:

Available Formats

Nutrition Community Era College Wastewater

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nutrition Community Era College Wastewater

Uploaded by

Copyright:

Available Formats

60 J. Rojas, T.

Zhelev / Computers and Chemical Engineering 38 (2012) 52–63

Dissolved oxygen (mg/l)

Aeration flowrate (vvh)

Dissolved oxygen (mg/l)

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Aeration flowrate (vvh)

Contents lists available at SciVerse ScienceDirect

Computers and Chemical Engineering

Energy efﬁciency optimisation of wastewater treatment: Study of ATAD

1. Introduction 2% of the total amount of electricity generated (Batts, Burton, &

Received: 1 July 2017; Accepted: 1 August 2017; Published: 9 August 2017

Appl. Sci. 2017, 7, 813; doi:10.3390/app7080813 www.mdpi.com/journal/applsci

Appl. Sci. 2017, 7, 813 14 of 21

Figure 9. Ammonium concentration in the effluent for selected controller designs.

Table 4. Performance indices variation respect to the BSM1 PI default strategy.

Controller % AE Variation % PE Variation % OCI Variation

Conflicts of Interest: The authors declare no conflict of interest.

Description Name Values

The equations of the mathematical model are the following:

3. Methodology Ti = qin xiin − qout xi (12)

Parameter Value/formula Note Source

tr = max{ts , tp }. (25) Two designs will be considered for optimisation: a single-stage

(a) (b) 2.2

Dissolved oxygen (mg/l)

Aeration flowrate (vvh)

Dissolved oxygen (mg/l)

0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Aeration flowrate (vvh)

You might also like