Design of a Sequencing Batch Reactor

Sequence with an Input Load Partition in a

Simulation-Based Experimental
Environment
Nadja Hvala, Mario Zec, Milenko Roš, Stanko Strmčnik

ABSTRACT: A study is presented that considers optimization of a because it may disrupt the system. Also, during process operation,
sequencing batch reactor (SBR) operational sequence. Optimization is many process variables are difficult to monitor because they can-
performed with the aid of a laboratory pilot plant and a process model. not be measured or are determined only by lengthy laboratory
The model was included in the study to facilitate a search for optimal
analyses.
operating conditions because performing experiments solely on the pilot
Because of the above properties, wastewater treatment repre-
plant would be rather extensive, expensive, and time-consuming. The
model used in the study is the well-known Activated Sludge Model No. sents an area where models can be usefully applied for different
1 with minor modifications. The optimization addresses a split-feed purposes. Models are used for better understanding processes, as a
operating mode with input load partitioned into two parts and with two tool in plant design and upgrading (for choosing the most suitable
successions of aerobic–anoxic phases. In the study, the duration of batch technology and plant dimensions), as an aid in operation (for
phases and the time of the second input addition are optimized so that determining the optimal operating point), and for better (model-
the desired effluent nitrogen concentration is achieved and the need for based) process control.
external carbon addition in the anoxic phase is reduced or eliminated. Despite the increasing use of models in this area, reliable models
When an optimal SBR sequence computed by the model was verified on
of wastewater treatment processes are still relatively difficult to
the pilot plant, better performance of the actual process was also
achieved, although some phenomena were observed that were not
design and apply. The reason is that wastewater treatment pro-
predicted by the model. Results of the study indicate that the available cesses are poorly defined systems, for which a valid model struc-
wastewater treatment plant models, although a simplification of reality, ture is not known and the model parameter values cannot be
can be usefully applied for process optimization. Water Environ. Res., accurately specified (Kops et al., 1997). Difficulties in modeling
73, 146 (2001). wastewater treatment processes lie in quantification of the various
KEYWORDS: sequencing batch reactor (SBR), wastewater treatment,
complex biochemical reactions involved, the difficulty in gathering
denitrification, split-feed operating mode, activated-sludge modeling, sim- all of the required identification data, the large number of param-
ulation. eters and state variables typically included in model structure,
model equations expressing high nonlinearities, and the fact that
not all model parameters can be uniquely identified from plant
Introduction measurements (Sperling, 1993). Because of all of these factors,
Modeling and simulation are important and widely used meth- models of wastewater treatment processes should be applied with
ods for analysis and design of complex dynamic systems. Espe- care, especially when used for predicting process behavior under
cially since the arrival of personal computers, they have spread different operating conditions.
throughout all branches of systems analysis and design and have In this paper, the use of an activated-sludge model for optimiz-
found a wide field of application. They are particularly useful in ing plant operating conditions is examined. The approach is based
cases where systems with the following properties are considered: on a mathematical process model, simulation, and optimization.
(a) systems that are not yet operating, and (b) operating systems in Because of difficulties in modeling wastewater treatment pro-
which the system is complex, with a large number of processes and cesses, the approach is also combined with experiments performed
complex interactions; the system has too great or too little time on the pilot plant. In particular, optimal operating conditions are
constants; and experiments can seriously interrupt or damage the first determined with the aid of the process model and afterwards
system. verified with experiments on the plant. Combining simulation
Wastewater treatment processes have many of the above prop- experiments performed with the aid of the model with actual plant
erties; they are complex systems consisting of several biological experiments has the following advantages: use of the model is
reactions in which different types of microorganisms degrade the much faster and easier (compared to long-lasting experiments on
polluted ingredients in the wastewater. The microorganisms adapt actual or pilot plants) and therefore decreases the time needed to
to changeable environmental conditions, and therefore express find a solution. In addition, it reduces design costs and, by a
time-varying behavior. The processes have different time constants systematic search of the entire operating region, ensures the opti-
that, together with complex interactions, make them difficult to mality of the final chosen solution. On the other hand, verification
analyze. Experimentation on an existing plant is quite demanding of the chosen solution on the actual plant helps to uncover any

146 Water Environment Research, Volume 73, Number 2

 Hvala et al.

problems resulting from model uncertainty or a poor predictive the C:N ratio in the anoxic phase must be approximately 3:1
capability of a model under new operating conditions. By com- (depending on the specific carbon source). Because, in the opera-
bining simulation and actual plant experiments, only the most tion of a SBR, soluble organic substrate is already consumed in the
promising solutions determined with the aid of the simulation preceding aerobic phase, different measures are typically applied
model are considered and examined in the actual plant. In this way, for securing enough soluble organics for denitrification.
the model assists in achieving an optimal specification of the One possibility for ensuring organic matter for denitrification is
necessary pilot-plant experiments. addition of external carbon in the anoxic phase (Abufayed and
In this paper, this approach is used to optimize the operational Schroeder, 1986; Aspegren et al., 1992; and Demuynck et al.,
sequence of a sequencing batch reactor (SBR) for wastewater 1994). This increases the plant’s operating costs and is, for eco-
treatment. The aim is to define the sequence and duration of batch nomical reasons, undesirable. Another possibility is use of an
phases so that the desired wastewater treatment quality is achieved intracellular electron donor (glycogen) that could be generated in
at the end of the batch. In the study, emphasis is placed on the aerobic phase if it is long enough (Alleman and Irvine, 1980,
optimization of nitrification and denitrification processes per- and Silverstein and Schroeder, 1983). In this case, increased en-
formed in the aerobic and anoxic phases of the batch cycle, ergy consumption is to be expected because of longer aeration.
respectively. These two phases should be optimized so that effi- Another possible way of ensuring a carbon source for denitri-
cient nitrogen reduction is achieved and the need for external fication involves using substrate from the influent. This means that
carbon addition, resulting from a low carbon-to-nitrogen (C:N) the aerobic phase is stopped before all soluble substrate is used.
ratio in the anoxic phase, is reduced or eliminated. To achieve this, For this purpose, some authors (Demuynck et al., 1994; Irvine and
a split-feed operating mode is examined with input load partitioned Ketchum, 1989; and Jones et al., 1990) have proposed performing
into two parts and with two successions of aerobic–anoxic phases several repetitions of shorter aerobic and anoxic phases instead of
in the SBR cycle. In the study, optimal batch-phase times were only one succession of longer aerobic and anoxic phases. Some
chosen so that efficient denitrification (and therefore the desired authors have also proposed saving influent COD for denitrification
effluent nitrogen concentration) would be achieved without the by influent load partitioning into two (or more) equal parts (ATV,
need for external carbon addition in the anoxic phase. The optimal 1997, and Kayser, 1995). This approach is known as a split-feed
batch-phase times were chosen with the aid of the process model operating mode and will be further investigated here.
and optimization, and the final solution was also tested on the
laboratory plant. Experimental Environment.
Optimal design of an operational sequence based on a split-feed
Sequencing Batch Reactor operating mode was performed with the aid of an experimental
The SBR is an activated-sludge system for wastewater treatment system that supports three primary steps of experimentation: plan-
operating on a fill-and-draw (batch) basis. Compared with contin- ning the experiment, conducting the experiment, and evaluating
uous wastewater treatment, it is a compact and relatively simple the experiment. As shown in Figure 1, during experimentation
system because all of the treatment processes take place in one these three steps are iteratively repeated after new knowledge is
reactor. Sequencing batch reactors can be used to remove organic gained about the system on the basis of the performed experiments
and nitrogen- and phosphorus-containing compounds from waste- and until the optimal solution that satisfies the primary require-
water (Imura et al., 1993). With the same plant volume, an SBR ments of experimentation is found. In this approach, improved
can treat greater flows and greater loads than continuous waste- knowledge about the system was gained by the interchange be-
water treatment plants (Orhon and Artan, 1994). Also, problems tween pilot-plant experiments performed on the laboratory pilot
with sludge settling that are present in many continuous treatment plant and simulation experiments performed with the aid of the
plants are not as common in SBRs. process model. In particular, some experiments were first per-
The operation of a SBR is flexible and can be adjusted to the formed on the laboratory pilot plant that generated the data for
current biomass state and input variations, variations that are model identification. Following this, the model was used to sim-
typical for municipal (or in many cases, municipal and industrial) ulate different operational scenarios and to find an optimal oper-
wastewater. By changing the fill–react rates (duration of phases) ational sequence with the aid of optimization. Finally, the optimal
and the sequence of batch phases, process operation can be opti- solution was verified by performing experiments on the laboratory
mized and adjusted to the current state. pilot plant. Typically, the results obtained initiated a new cycle
A typical SBR cycle consists of several phases, namely filling, with improvements of the model (e.g., improved model structure
reaction (anaerobic, aerobic, and anoxic phases), settling, and
withdrawing. When only removal of organic matter and nitrogen
reduction are considered, three processes have to be accomplished:
chemical oxygen demand (COD) removal, nitrification, and deni-
trification.
The COD removal and nitrification are performed in the aerobic
phase and denitrification takes place in the succeeding anoxic
phase. For fast and efficient denitrification, two primary conditions
must be fulfilled: there have to be enough nitrates and enough
soluble organic substrate in the anoxic phase. To obtain enough
nitrates, an efficient nitrification (conversion of ammonium to
nitrates) has to be performed before denitrification (Coelho et al.,
2000). In addition, in the literature (Henze et al., 1997, and Orhon
and Artan, 1994) it is reported that for a typical denitrification rate Figure 1—Iterative steps in experimental design.

March/April 2001 147

Hvala et al.

or better estimates of parameter values) and a search for more control system to facilitate experimentation under various operat-
optimal solutions, or additional pilot-plant experiments. ing conditions and the interchange between the simulation and
In this experimental environment, the system used for conduct- actual plant experiments (Šubelj et al., 1997).
ing the experiments is either a laboratory pilot plant or a simulation First, all of the process data; that is, online measurements,
program and corresponding mathematical model. The inclusion of laboratory analyses, and batch parameters, are collected in a uni-
the mathematical process model into the experimental environ- fied process database. This enables easy access to process data
ment helps the gain of additional knowledge about process behav- needed for the process model design and for evaluation of the
ior. This is attributed to the fact that performing experiments by experiments performed on the pilot plant.
simulation allows testing of different operational scenarios in a Second, the online control system is based on flexible recipe
systematic way. It is also much faster, cheaper, and easier. Hence, control, which allows for desired variability and easy adjustment
more experiments in a wide range of operating conditions can be of the batch recipe parameters (e.g., sequence and duration of
easily performed. In addition, in the pilot-plant experiments not all phases) in an automatically operated control system. This is im-
variables that are present in the model could be measured. Thus, portant during experimentation when different technological solu-
accessing these variables, which would otherwise not be observed, tions are tested on the pilot plant.
may enhance understanding of process behavior. Last, with the help of specially designed functions, real-time
Laboratory Pilot Plant. The SBR laboratory pilot plant used in data exchange is possible between the online process database and
the study consists of a 73-L rectangular reactor (Figure 2). Air is the control design software tool Matlab (The MathWorks, Inc.,
blown into the reactor through valves V1 and V2 and a rubber Natick, Massachusetts). This is especially important for the design
membrane. Valves V3 and V4 are used for discharging the treated of online control algorithms because algorithms that are designed
water and valve V5 is meant for excess sludge discharge. Opera- in a simulation environment and tested with the aid of a mathe-
tion of the pilot plant is monitored by five online measurements, matical model can also be immediately applied in real-time pilot-
namely pH, oxidation–reduction potential (ORP), dissolved oxy- plant control.
gen (DO) concentration, temperature (T), and level (volume). Simulation Model. Simulation experiments were performed
Pilot-plant operation is controlled by a Mitsubishi (Tokyo, Japan) with the aid of the mathematical process model further explained
AnS programmable logic controller (PLC), which is connected to below. To represent the biological reactions in a SBR, a Mantis
the supervisory system. The latter is based on a Windows NT 3.51 model was used (Hydromantis, 1997). This model is identical to
workstation (166 mHz Pentium personal computer) with Factory the well-known Activated Sludge Model No. 1 (Henze et al.,
Link (US Data, Richardson, Texas) supervisory control and data 1987), except that it contains two additional growth processes for
acquisition (SCADA) software. autotrophic and heterotrophic bacteria during conditions of low
Several additional functions were added to the supervisory ammonia and high nitrate concentration and includes temperature
dependence of the kinetic parameters. A one-dimensional settling
model (Takacs et al., 1991) was used for modeling the clarification
and thickening processes in the settling phase. Simulations were
performed using GPS-X simulation software (Hydromantis, 1997),
which is produced especially for simulation of wastewater treat-
ment processes.

Sequencing Batch Reactor Cycle Optimization

The SBR cycle optimization was performed based on pilot-plant
and simulation experiments. The procedure consisted of the fol-
lowing three steps: (1) pilot-plant measurements of the one-fill
cycle, (2) simulation model adjustment (estimation of initial states
and model parameters), and (3) simulation and optimization of the
two-fill cycle.
Pilot-Plant Measurements of the One-Fill Cycle. Initial pilot-
plant measurements were performed to adjust the simulation
model to the SBR pilot-plant conditions. For this purpose, pilot-
plant operation was monitored by the already mentioned online
measurements and some laboratory measurements, namely COD,
ammonium-nitrogen (NH4–N), nitrite-nitrogen (NO2–N), nitrate-
nitrogen (NO3–N), total Kjeldahl nitrogen (TKN), soluble phos-
phorus (PO4–P), and mixed liquor suspended solids (MLSS).
Measurements were performed during one batch. In the experi-
ment, the volume of the activated sludge remaining in the reactor
from the previous batch was 45.6 L and the reactor received 10 L
of wastewater. Input wastewater was a mixture of peptone, soluble
phosphorus, and water, with the characteristics shown in Table 1.
After the addition of wastewater, the MLSS concentration in the
reactor was 3.6 g/L. The sludge volume index (SVI) was 43 mL/g.
Figure 2—Laboratory SBR pilot plant with computer con- As shown in Table 2, the SBR cycle consisted of seven phases with
trol system. a total duration of 480 minutes. In the cycle, one succession of

148 Water Environment Research, Volume 73, Number 2

 Hvala et al.

Table 1—Input wastewater composition for the batch

cycle used for initial states and parameter estimation.

Parameter Concentration

COD, mg/L 906

TKN, mg/L 107
NH4–N, mg/L 6.9
NO2–N, mg/L ⬍0.05
NO3–N, mg/L 3.1
Total nitrogen, mg/L 110
PO4–P, mg/L 8.1

aerobic and anoxic phases was performed. During the experimen-

tal period, sludge was wasted once a day before settling from the
mixed reactor. Approximately 4 L of sludge was wasted manually Figure 3—Measured (points) and simulated (lines) data
to maintain the MLSS concentration in the reactor (during the react of the SBR cycle (NO2–N, NO3–N, and NH4–N).
phases) at approximately 3.5 g/L. This means that the solids
retention time (SRT) was approximately 14 days. Loading can be
calculated as 0.13 kg COD/kg MLSS䡠d.
go into a true anoxic phase. The model predicts that the transition
Simulation Model Adjustment. The SBR model was adjusted
from the aerobic to anoxic state occurs immediately after the
to the pilot-plant conditions by estimating initial states and model
concentration of oxygen drops to zero. Thus, some deviations exist
parameters. This was done by adjusting their values so that good
between the model and the process that could not be explained
agreement was achieved between the measured and simulated
from the experiment. Further comments on this problem are given
process variables. The adjustment was performed manually in
in the Discussion section. The only way to simulate this kind of
several cycles by using prior knowledge (gathered by simulation)
delay in the aerobic–anoxic transition was to extend the aeration
of how a single parameter or initial state, or a combination of
time in the model and simulate a quasiaerobic state that was
parameters and initial states, influences the process state variables.
evident from the measurements. With prolonged aeration, the
During this process, emphasis was placed on estimating initial
measured and simulated data agree well in the entire cycle, as seen
concentrations in the reactor because many initial solids concen-
in Figure 5.
trations, such as the ratio between heterotrophic and autotrophic
Simulation and Optimization of the Two-Fill Cycle. The
biomass, slowly biodegradable substrate, particulate organic inert
above simulation model was used to optimize the SBR cycle. The
compounds, particulate biodegradable organic nitrogen, and so
goal was to reduce effluent nitrate concentration without extra
forth, were not accurately known. Finally, the adjustment to pilot-
COD supply in the anoxic phase, while also preserving a low
plant conditions was achieved by changing the initial states only,
effluent ammonium concentration. This was expected to be
whereas stoichiometric and kinetic parameters were set to values
achieved by dividing the batch influent into two parts and by
as recommended in Activated Sludge Model No.1 (typical param-
performing two successions instead of only one succession of
eter values at neutral pH) (Henze et al., 1987).
aerobic–anoxic phases. With the second influent added in the first
Figures 3 and 4 show the fit between the simulated and mea-
anoxic phase, an increased concentration of soluble organics in this
sured data of the SBR cycle. As can be seen, the simulated data
phase is achieved, which improves denitrification and thus reduces
agree quite well with measurements in the first part of the batch.
the effluent nitrate concentration. The principle of adding succes-
However, important deviations can be observed at the beginning of
sions of aerobic–anoxic phases is the same for multiple divisions
the anoxic phase. In this part, the measured data show that even
as well. For example, four successions of aerobic–anoxic phases
after the concentration of oxygen drops to zero (in the 310th
are required for influent partitioning into four parts.
minute of the measured SBR cycle), the reactor remains aerobic
The aim of optimization of a two-fill cycle was to determine the
(nitrification continues, ammonium remains stable). Only at ap-
optimal duration of phases in the cycle. For this purpose the
proximately the 360th minute of the cycle does the reactor finally
following decisions were made. The duration of the filling–anoxic–
anaerobic phase was shortened from 45 to 35 minutes. This deci-
Table 2—Phases in the SBR cycle used for initial states sion was made because in the two-fill cycle, it is expected that a
and parameter estimation. smaller amount of nitrates needs to be denitrified in the filling
phase (because of expected better performance, fewer nitrates
Phases Duration, minutes should remain in the reactor from the previous batch). Conse-
quently, by shortening this phase, less influent COD is used in
Filling–anoxic–anaerobic 45 anaerobic conditions and is thus saved for succeeding phases.
Aerobic 215 Second, the times of the stripping, settling, withdrawing, and
Anoxic 160 idling phases were set the same as in the one-fill cycle. Last, the
Stripping 15 times of the remaining four phases (first aerobic, first anoxic,
Settling 15 second aerobic, and second anoxic) should be determined in an
Withdrawing 10
optimal way to meet the desired design requirements.
Idling 20
Considering all of the above, the only times that needed to be

March/April 2001 149

Hvala et al.

Step 3: The second influent addition was added in the first

anoxic phase. Because of the already mentioned problem of aer-
obic delay, the time for the second filling was set at 52 minutes
after the start of the first anoxic phase, to ensure that the second fill
was added under anoxic conditions.
Step 4: The duration of the first anoxic phase was set long
enough for complete denitrification in this phase.
Step 5: Finally, the times of the second aerobic and anoxic
phases were determined based on an optimization performed using
an optimization algorithm. Because of requirement d, the total time
of the second aerobic and anoxic phases was constrained to 240
minutes (computed from the total batch time and the already
determined times of batch phases), whereas the time of the second
aerobic phase should be, because of requirement c, less than or
equal to 155 minutes. In the optimization, the following objective
Figure 4 —Measured and simulated data of the SBR cy- function F was minimized
cle (DO and ORP).
F⫽ 冘 共␣t AER ⫹ ␤fCOD ⫹ ␥fNH4⫺N ⫹ ␦fNO2⫹NO3⫺N ⫹ efSTKN兲 (1)

Where
determined were the times of the aerobic and anoxic phases. These tAER ⫽ time of the second aerobic phase;
were determined in a heuristic procedure where the process model,
simulation, and optimization were used. In the procedure, the ␣,␤,␥,␦,e ⫽ constants representing weights of different
following requirements were taken into account: (a) The concen-
criteria in the objective function and were set to 1; and
tration of nitrates after the first aerobic phase must be equal to a
nitrate concentration that could be denitrified in the following fCOD, fNH4 ⫺ N, fNO2 ⫹ NO3 ⫺ N, and fSTKN (soluble TKN) ⫽ measure-
anoxic phase with COD added in the second fill. (b) The second ments of the violation of effluent concentrations, Xeff, over the
filling must be applied when the reactor is in the anoxic phase. (c) desired values kX, determined as follows:
The total aeration time must be less than or equal to the aeration
time in the one-fill cycle. (d) The total duration of the two-fill cycle X eff ⫺ kX ⫹ 兩Xeff ⫺ kX兩
must be equal to that of the one-fill cycle. fX ⫽ (2)
2
The final procedure for determining the times of the aerobic and
anoxic phases consisted of the following five steps. where kCOD ⫽ 60 mg/L, kNH4 ⫺ N ⫽ 1 mg/L, kNO2 ⫹ NO3 ⫺ N ⫽ 4 mg/L,
Step 1: The amount of nitrates that could be denitrified in the and kSTKN ⫽5 mg/L.
first anoxic phase was determined by taking into account the A complete optimal two-fill cycle with nine phases, and their
influent COD added in the second fill and the observed denitrifi- duration times obtained after optimization, are listed in Table 3.
cation rate. The result was the minimum nitrate concentration
needed at the end of the first aerobic phase. Verification of the Optimal Two-Fill Cycle on the Pilot
Step 2: The duration of the first aerobic phase was set long Plant
enough so that nitrification, performed in the aerobic phase, could The above-specified optimal two-fill cycle was verified by sim-
produce enough nitrates to be denitrified in the first anoxic phase. ulation and in a pilot plant. In the pilot-plant experiments, the
adaptation of the process to the new operating conditions took 10
days (30 cycles). After the adaptation period, measurements were
performed for one batch. The composition of input wastewater
used for testing the optimal two-fill cycle is shown in Table 4.
Wastewater component concentrations were similar to those used
in the cycle for parameter estimation (Table 1). In the experiment,

Table 3—Optimal two-fill cycle.

Phases Duration, minutes

Filling–anoxic–anaerobic 35
Aerobic I 60
Anoxic I 85
Aerobic II 80
Anoxic II 160
Stripping 15
Settling 15
Figure 5—Measured (points) and simulated (lines) data
Withdrawing 10
of the SBR cycle (NO2–N, NO3–N, and NH4–N) with pro-
Idling 20
longed aeration.

150 Water Environment Research, Volume 73, Number 2

 Hvala et al.

Table 4 —Input wastewater composition for optimal two- Table 6 —Total nitrogen reduction in the one-fill cycle
fill cycle experiment. and optimal two-fill cycle.

Parameter Concentration Total nitrogen One-fill Optimal two-fill

reduction cycle cycle
COD, mg/L 925
TKN, mg/L 110 Simulation 87% 93%
NH4–N, mg/L 7 Pilot plant 82% 92%
NO2–N, mg/L 0
NO3–N, mg/L 5.2
Total nitrogen, mg/L 115 tions. Therefore, an important point to discuss is the agreement
PO4–P, mg/L 8.7 between the simulation model and the pilot-plant behavior.
As already mentioned, deviations between the model and the
process were observed in the one-fill cycle after aeration was
the MLSS concentration in the reactor was 3.7 g/L and SVI was stopped. In this case, the pilot-plant experiments show a delay in
116 mL/g. the aerobic–anoxic transition, and the simulation model used in the
Figures 6 and 7 show the results of simulation and pilot-plant study does not anticipate this kind of behavior in its structure or
experiments performed with the optimal two-fill cycle. By first processes. Therefore, the only way to simulate the observed phe-
examining the efficiency of the designed two-fill cycle in relation nomenon was to extend the aerobic phase. Most likely, the delay
to the effluent quality requirements, it can be seen that in both the was caused by the weak oxygen uptake rate at the end of the
simulation and pilot-plant experiments, considerable reduction of aeration period, when soluble organics and ammonium are already
the final nitrate concentration is achieved compared with that of consumed and only endogenous respiration is taking place. This
the one-fill cycle. Close inspection also shows that in the first weak oxygen uptake rate is not enough to take up all of the oxygen
anoxic phase, a complete denitrification of nitrates (released in the present in the system because of oxygen transfer from the atmo-
preceding aerobic phase) is achieved as a result of the second input sphere. The presence of oxygen in the reactor is also evident from
addition in this phase. Experiments also show that in the two-fill the measured DO concentration, which, after aeration is stopped
cycle, the ammonium concentration increased slightly compared (Figure 4), decreases at a slower rate than predicted by the model.
with the one-fill cycle, but in both simulation and pilot-plant It is probable that even more oxygen was present in the reactor
experiments it is less than 1.5 mg/L. because of the emptying of the aeration system (after the air supply
A summary of results achieved is also provided in Tables 5 and valve is closed, some time is needed for emptying the rubber
6. From Table 6 it can be seen that in the two-fill cycle experiment membrane and all of the pipes), but was not measured because of
performed on the pilot plant, a 92% reduction of total nitrogen is the probe position. A similar case can also be observed in Figure
achieved, which is 10% more than in the one-fill cycle. From this 6, where a much lower simulated DO concentration is obtained at
it can be concluded that pilot-plant verification of the designed the end of the batch than is measured in the pilot plant. This was
two-fill cycle confirms an improved process operation in relation because the reactor was in the settling and withdrawing phases and
to the effluent quality, which was the goal of optimization. its contents were not mixed. Because of its position, the pilot-plant
DO sensor shows the DO concentration of the upper (clean) part of
Discussion the reactor, whereas the simulated data represent the DO concen-
In the study, a search for an improved SBR operating sequence tration at the discharge point, which is, according to Figure 2, in
was performed with the aid of a mathematical process model and the middle of the reactor. However, the DO concentration in this
its prediction of process behavior under different operating condi- part of the SBR cycle is not too important for overall SBR
operation.

Figure 6 —Measured and simulated data of the optimal Figure 7—Measured (points) and simulated (lines) data
two-fill cycle (DO and ORP); ON/OFF: aeration on– off. of the optimal two-fill cycle (NO2–N, NO3–N, and NH4–N).

March/April 2001 151

Hvala et al.

Table 5—Final effluent concentrations for one-fill and two-fill cycles in simulation and pilot-plant experiments.

COD, TKN, NH4–N, NO2 ⴙ NO3–N, Total nitrogen,

Experiment mg/L mg/L mg/L mg/L mg/L

One-fill cycle (simulation) 48 2 0.2 11.8 13.8

One-fill cycle (pilot plant) 33 5 0.5 14.7 19.7
Optimal two-fill cycle (simulation) 19 4 1.5 4.2 8.2
Optimal two-fill cycle (pilot plant) 41 7 1 1.8 8.8

In those cases where there were some nutrients in the reactor Namely, this study is based on a defined and constant influent. In
(which cause a greater oxygen uptake rate), a delay of aerobic– actual plant operation, however, a variable input load occurs. In
anoxic conditions was not noted and oxygen immediately dropped most cases, influent volumes are stabilized using an equalization
to zero after the aeration was stopped (see Figure 6). basin, but the concentrations still vary from batch to batch. Al-
The quality of the agreement between the measured and simu- though experiments have not yet been performed with variable
lated process behavior is also important in the two-fill cycle, as in influent concentrations, some basic directions of the expected
this case, the model predicts the process behavior under different behavior and control of the SBR cycle under such conditions may
operating conditions and therefore shows the predictive capability be indicated. One possible way to attain an optimal two-fill cycle
of the model. From the results shown in Figure 7 it can be seen that (or optimal one-fill cycle) under variable influent is to measure (or
also in this case, a relatively good agreement between the simu- estimate) influent load, perform online optimization, and change
lated and measured data was obtained. Some greater deviations can the duration of the phases accordingly. This procedure requires
be seen in the second part of the cycle, where measured NH4–N sophisticated sensors, optimization routines, and actuators, and
and NO3–N concentrations differ from the simulated values. This therefore has limited applicability in practice. Today, SBRs with
could be explained by the adaptation of the process to the new variable input loads are most often operated with longer-than-
operating conditions. That is, in the pilot-plant experiments several needed durations (for an average load) of batch phases and COD
cycles were performed to adapt the biomass to the new operating addition. The same principles could also be applied to the speci-
conditions. Similarly, in the simulation experiments, the adaptation fication of an SBR cycle with influent load partitioning: a longer-
was simulated by running the simulation model several cycles until than-needed aerobic phase will produce greater nitrate concentra-
the process reached a steady state. The latter means that the tions to be denitrified in the succeeding anoxic phase, and a
process behavior, with a constant input load imposed on it, did not longer-than-needed anoxic phase will ensure that the available
change much from batch to batch. It is likely that not all of the COD in the second (and third, etc.) fill is used in an anoxic
adaptation processes that take place in the actual system are also environment. To summarize, even in cases of variable input load
well represented in the simulation model. Therefore, the simulated and fixed durations of phases, effluent NO3–N concentrations can
steady state was probably different from that of the pilot plant. For be reduced by influent load partitioning.
example, in the two-fill cycle, better denitrification conditions
(longer anoxic phases, more COD, and hence a more favorable Conclusions
C:N ratio) stimulate the growth of heterotrophic biomass and A study was presented here in which wastewater treatment plant
reduce the growth of autotrophic biomass. If this is taken into operating conditions were optimized with the aid of an experimen-
account and the initial heterotrophic-to-autotrophic biomass ratio
is increased, a much better agreement between the simulated
variables and pilot-plant measurements can then be achieved (see
Figure 8).
Another important phenomenon, which was observed at the
pilot plant but was not predicted by the model, is an increase of
SVI in the two-fill cycle. During biomass adaptation to the two-fill
cycle conditions, the SVI increased from 43 mL/g in the one-fill
cycle to 116 mL/g in the two-fill cycle. The reasons for an increase
in SVI need to be further investigated. For example, in the liter-
ature (Arora et al., 1985; Chiesa and Irvine, 1982; and Irvine and
Ketchum, 1989) it is reported that the substrate concentration and
the mixing and aeration strategy during fill have an effect on SVI.
The problem with increased SVI indicates that the part of the
simulation model responsible for representing the settling in the
SBR should be improved. In addition, the SVI should be consid-
ered in the cost function when the SBR cycle is optimized. After
the completion of two-fill cycle experiments, the pilot plant was
switched back to the one-fill cycle and the SVI decreased quickly. Figure 8 —Measured and simulated data (increased het-
An open question arising from this study also addresses the erotrophic-to-autotrophic biomass ratio) of the optimal
two-fill cycle performance in relation to treatment efficiency and two-fill cycle (NO2–N, NO3–N, and NH4–N); ON/OFF: aer-
the applicability of the results obtained to variable input load. ation on– off.

152 Water Environment Research, Volume 73, Number 2

 Hvala et al.

tal system. The latter consisted of a laboratory pilot plant and its 2001.
mathematical model. The model was included in the experimental
environment because searching for optimal operating conditions is References
Abufayed, A.A., and Schroeder, E.D. (1986) Kinetics and Stoichiometry of
typically an optimization problem, where the desired goals should
SBR/Denitrification with a Primary Sludge Carbon Source. J. Water
be reached through a change in operating parameters. In the search
Pollut. Control Fed., 58, 398.
process several possible solutions are examined, which are much Abwassertechnische Vereinigung (ATV) (1997) Belebungsanlagen mit
easier to perform with the aid of a simulation model than by Aufstaubetrieb. Gesellschaft zur Forderung der Abwassertechnik,
extensive experimentation in a pilot plant. Hennef, Ger.
The case study presented refers to a SBR operation where batch Alleman, J.E., and Irvine, R.L. (1980) Storage-Induced Denitrification
times of a two-fill cycle were optimized so that the desired effluent Using Sequencing Batch Reactor Operation. Water Res. (G.B.), 14,
quality was approached without extra COD supply in the anoxic 1483.
phase. In the study, the optimal cycle was first designed with the Arora, M.L.; Barth, E.F.; and Umphres, M.B. (1985) Technology Evalu-
aid of the model and then verified by pilot-plant experiments. ation of Sequencing Batch Reactors. J. Water Pollut. Control Fed.,
57, 867.
A comparison of the results obtained by simulation and in the
Aspegren, H.; Andersson, B.; Nyberg, U.; AND Jansen, J. la C. (1992)
pilot plant shows that optimal operating conditions determined
Model and Sensor Based Optimization of Nitrogen Removal at Klag-
with the aid of the model also give better process performance in shamn Wastewater Treatment Plant. Water Sci. Technol., 26, 1315.
an actual process. Therefore, it may be concluded that the model Chiesa, S.C., and Irvine, R.L. (1982) Growth and Control of Filamentous
was able to predict the correct manner to change process operating Microbes in Activated Sludge—An Integrated Hypothesis. Paper pre-
conditions for improved process performance. The simulation re- sented 55th Annu. Water Pollut. Control Fed. Conf. Exposition, St.
sults proved to be comparable to pilot-plant experiments, and the Louis, Mo.
simulation study gave a good indication of what might be expected Coelho, M.A.Z.; Russo, C.; and Araujo, O.Q.F. (2000) Optimization of a
from the process under new operating conditions. Sequencing Batch Reactor for Biological Nitrogen Removal. Water
Pilot-plant operation under new conditions also revealed some Res. (G.B.), 34, 2809.
Demuynck, C.; Vanrolleghem, P.; Mingenau, C.; Liessens, J.; and Ver-
phenomena that were not predicted by the model. This is because,
straete, W. (1994) NDBEPR Process Optimization in SBRs: Reduc-
although a rather complex model was used, it was still a simpli-
tion of External Carbon-source and Oxygen Supply. Water Sci. Tech-
fication of processes that take place in an actual system. The use of nol., 30, 169.
simpler models is preferred over more complex ones because less Henze, M.; Grady, C.P.L. Jr.; Gujer, W.; Marais, G.v.R.; and Matsuo, T.
effort is needed for estimating model parameters. (1987) Activated Sludge Model No. 1. Int. Assoc. Water Pollut. Res.
Overall, the study presented suggests that available wastewater Control, London.
treatment plant models, although simplifications of reality, can be Henze, M.; Harremoës, P.; LaCour Jansen, J.; and Arvin, E. (1997)
usefully applied for process optimization. Although not all phe- Wastewater Treatment: Biological and Chemical Processes. Springer,
nomena could be observed by the model, good predictions can still Heidelberg, Ger.
be made, which suffice in a search for better process operation. Hydromantis, Inc. (1997) GPS-X - Technical Reference. Ontario, Can.
Imura, M.; Suzuki, E.; Kitao, T.; and Iwai, S. (1993) Advanced Treatment
The observed model deficiencies suggest that process optimization
of Domestic Wastewater Using Sequencing Batch Reactor Activated
performed with the aid of a model should be combined with actual
Sludge Process. Water Sci. Technol., 28, 267.
plant experiments, so that eventual unmodeled effects expressed Irvine, R.L., and Ketchum, L.H., Jr. (1989) Sequencing Batch Reactors for
under new operating conditions can be noticed. Biological Wastewater Treatment. CRC Crit. Rev. Environ. Control,
Future research will address the search for optimal operating 18, 255.
conditions by taking into account the observed phenomena (e.g., Jones, W.L.; Schroeder, E.D.; and Wilderer, P.A. (1990) Denitrification in
delayed aerobic conditions, SVI increase) and the fact that the a Batch Wastewater Treatment System Using Sequestered Organic
model does not describe them. Research into the optimal cycle in Substances. J. Water Pollut. Control Fed., 62, 259.
the case of input load variations will also be addressed. Kayser, R. (1995) Bemessung von Belebungsanlagen mit Aufstaubetrieb
(SBR-Anlagen). Wasser-Abwasser (Ger.), 136, 625.
Kops, S.; Vangheluwe, H.; Claeys, F.; and Coen, F. (1997) The Process of
Acknowledgments
Model Building and Simulation of Ill-Defined Systems: Application
Credits. The support of the Ministry of Science and Technology to Wastewater Treatment. Proc. Int. Assoc. Math. Comput. Simu.
of the Republic of Slovenia, Ljubljana, is gratefully acknowledged. Symp. Math. Modelling 2nd MATHMOD, Vienna, Aus., 859.
Authors. N. Hvala is a research associate and S. Strmčnik is the Orhon, D., and Artan, N. (1994) Modelling of Activated Sludge Systems.
head of the Department of Computer Automation and Control at Technomic Publishing Company, Lancaster, Pa.
the J. Stefan Institute in Ljubljana, Slovenia. At the time this work Silverstein, J., and Schroeder, E.D. (1983) Performance of SBR Activated
was performed, M. Zec was a Ph.D. student, and is now employed Sludge Processes with Nitrification/Denitrification. J. Water Pollut.
at the Ministry of Environment and Physical Planning in Ljubljana. Control Fed., 55, 377.
M. Roš is a professor at the National Institute of Chemistry, Sperling, M.V. (1993) Parameter Estimation and Sensitivity Analysis of an
Ljubljana. Activated Sludge Model Using Monte Carlo Simulation and the
Analyst’s Involvement. Water Sci. Technol., 28, 11-12, 219.
Correspondence should be addressed to Nadja Hvala, J. Stefan
Šubelj, M.; Hvala, N.; Godena, G.; and Zec, M. (1997) Process-Oriented
Institute, Jamova 39, SI-1000, Ljubljana, Slovenia. Development of Computer Aided Experimental Environment. Paper
Submitted for publication August 25, 1998; revised manuscript presented 7th Int. Workshop: Instrum., Control Auto. Water Waste-
submitted October 12, 2000; accepted for publication October 27, water Treat. Transport Syst., Brighton, U.K., 498.
2000. Takacs, I.; Patry, G.G.; and Nolasco, D. (1991) A Dynamic Model of the
The deadline to submit Discussions of this paper is July 15, Clarification-Thickening Process. Water Res. (G.B.), 25, 1263.

March/April 2001 153