A practical protocol for dynamic modelling of activated

sludge systems

Abstract Use of dynamic simulation models has become standard practice in The Netherlands. Since the
introduction around 5 years ago more then 100 full scale wastewater treatment plants have been modelled.
Initially very different approaches have been used varying in calibration approach, amount of sampling and
time investment. Based on the accumulated practical experiences the Dutch Foundation of Applied Water
Research STOWA has stimulated the development of a protocol to aid in the set-up and calibration of
models for full scale wastewater treatment plants. Herein the aim was to develop a protocol, which in
practice was easy to use, minimising time and costs effort, but give a reliable and useable method. In this
paper this protocol is briefly described.
Keywords Activated sludge; ASM calibration; dynamic simulation; modelling; protocol

Introduction
In order to stimulate utilisation of dynamic simulations of activated sludge systems, the
Dutch Foundation of Applied Water Research (STOWA) advised in 1995 to use the simula-
tion software SIMBA®. The use of one simulation package, which could quickly be intro-
duced to a great number of users, has greatly enhanced the introduction and acceptance of
dynamic modelling in The Netherlands. Since 1995 more then 100 full-scale wastewater
treatment plants (WWTPs) have been dynamic modelled by different Water Boards, with
or without the support of consultants. With the introduction of simulation tools also the
need for a better standardisation and some form of quality control became apparent. This
led to the development of a standard methodology for characterising the influent
(Roeleveld and Kruit, 1998) and for developing a calibrated model. In both cases, easy to
use practical methods were preferred above scientific exactness. Originally the simulation
studies started with a long (1–4 weeks) intensive measurement campaign collecting and
analysing hundreds of samples, and even potentially checking many of the model parame-
ters by respirometry or other tests. This is an expensive and long lasting procedure. It can be
shown that many parameters in full-scale systems are hardly sensitive (although they might
be in e.g. a batch experiment!) and therefore experimental evaluation of these parameters is
not useful. The same is true for some of the measured concentrations. Development of the
model as much as possible on historical data and well directed measurements save much
time and costs and make the use of modelling in everyday practice more feasible. When an
initial model is developed it can be used to evaluate which concentrations should be
measured in detail based on the sensitivity towards model parameters or variation over the
day. In this stage the readily available flow variations are already giving a sufficient
dynamic simulation.
In this paper, the guidelines for dynamic modelling of activated sludge systems as pro-
posed by STOWA are given. The objective was to provide a univocal protocol that will give
a standardised and familiar transfer of knowledge. The main purpose of the protocol was to 127
 reflect the present experience and knowledge in a structured way so that the simulation
models are used in such a manner that quality control becomes possible. A more
standardised use of dynamic simulations is essential because it is more and more used for
optimisation studies and for studying different aspects in the design phase of WWTPs. Not
only the quality control but also the planning should therefore be improved. The amount of
sampling and testing should be kept at a good but minimal level to allow a cost-effective
use of models. An exemplary use of the proposed scheme can be found in a paper by Meijer
J. J.W. Hulsbeek et al.

et al. (2001a,b).

Inventory
In order to develop the protocol practitioners with extensive experience were interviewed
for obtaining an overview on the presently applied methods (protocol structures) for using
ASM 1 for dynamic simulation of full scale wwtp’s. There was a special focus on the
aspects of influent characterisation, model structure and calibration.

Influent characterisation
Characterisation of the influent is very important in the application of the ASM models.
The characteristics were commonly determined according to the STOWA protocol
(Roeleveld, 2001; STOWA, 1996; STOWA, 1999) based on filter methods for characteris-
ing the slowly degradable COD and a BOD measurement for characterising the inert COD.
Practical experience showed that the influent characterisation gave no major problems.
Although one can discuss whether the filter methods are accurate, in practice the full-scale
simulation models are not too sensitive for the division between slowly and readily
degradable COD. Only the estimation of the inert COD from a BOD test posed problems. It
is needed to have the BOD20 due to the large dilution needed for such a BOD test method
proved unreliable. By measuring the BOD1,2,3,4,5,7,10 (Roeleveld and Kruit, 1998) and
essentially using the model to describe the BOD as function of the incubation time, the
method becomes reliable. Nevertheless it also become laborious and still sensitive to ana-
lytical errors. Also it seems from detailed calibrations that the fraction of inert COD in the
influent depends on the sludge age of the wwtp. Because of all the uncertainties we propose
below to use the fraction of inert particulate COD in the influent as a main calibration
factor.
A bottleneck was the uncertainty about the required monitoring frequency and duration
of the monitoring for characterisation of the different water and sludge flows. From the
inventory it became clear that for the Dutch situation 24-h flow proportional samplings
were sufficient for most purposes. If storm water events, or the influence of discontinuous
sludge handling need to be modelled, more frequent sampling is necessary. This also
applies if a fully correct modelling of the peak concentrations of NH4+-N and NO3–-N in the
effluent is desired. The duration of the measurement campaign necessary for the develop-
ment of a well working model depends on the accuracy of the required results. In general
1–3 days is considered sufficient for normal cases, whereas for the development of models
used in optimisation studies and for control strategies a measuring campaign is advised of 3
to 7 days and at least 7 days, respectively.

Model structure
Apparently it is quite easy to set up a model of an existing wwtp, but it has turned out to be
essential to visit the plant under study and evaluate it critically. Quite often the operators of
the plant have modified the process configuration slightly (or drastically), so that the sup-
posed operational conditions no longer hold true. Often errors in flows or control set points
128 prove more sensitive to the simulation model then the model parameters themselves
(Meijer et al., 2001b). In setting up the hydraulics of tanks there is great uncertainty in the
choice of the right number of compartments and their configuration in the model. For a
“plug flow” pre-denitrification tank, two compartments have shown to be sufficient. On
the other hand, for an oxidation-ditch type circuit (often applied in The Netherlands), the
number of compartments may be well over ten. This number even may become higher if a
vertical oxygen gradient needs to be incorporated (e.g. when surface aerators are applied).
For modelling process control it is important that the control structure is implemented in

J. J.W. Hulsbeek et al.

such a way that the process dynamics are described in an adequate way. It is not necessary
that the control algorithm is the same as in practice, one could e.g. enforce the DO in the
aeration tank to simulate the measured DO. Only if process control is the main subject of
study, is it required that the control structure is in accordance with the behaviour of the
process control in practice.

Calibration and validation

It appeared that in practice the choice of the kinetic and stoichiometric parameters changed
for calibration purposes was highly variable. This generally depended on individual
choices. After consultation, it has been considered best to calibrate the model in a stepwise
procedure: (i) sludge composition and production, (ii) nitrification and (iii) denitrification.
Parameters to be preferentially adapted were identified. Quite often it turned out that the
duration of the calibration was considered too long or longer then expected. This was
mainly caused by the fact that often initially the original plant data were directly used in the
simulations. During calibration the model had then often to be adapted or new measure-
ments needed to be performed. Therefore it is emphasised to put extra attention on the
set-up of the plant model before starting a detailed measurement and calibration
programme. With regard to the validation, sometimes data were used from a period highly
comparable with the period from which data for input and calibration were obtained (e.g. 1
week later). Of course, this is not correct: data from a clearly different period (e.g. summer
and winter conditions) should be preferably used for validation.

The protocol
Main structure
The main structure of the protocol is shown in Figure 1. The protocol is based on the results
of the inventory of existing protocol structures. In the next paragraphs the protocol struc-
ture is further detailed per phase. In the Figures also the required time for the different phas-
es is mentioned, this time is based on the time used by users who had at least calibrated one
WWTP model before.
The protocol structure is dependent on the objectives of the study. In the protocol three
different model studies are distinguished:
• System choice during the basic design;
• Optimisation studies for existing WWTPs;
• Development of control strategies for existent and new WWTPs.
The required level of specification of the model and the necessary monitoring frequency
of relevant flows is dependent on the objectives mentioned above. For the development of
control strategies the level of specification must be highest.

Process description
When the objectives of the study are clear, a definition of the relevant process components
can be made. Often it is not required to model the complete WWTP. Only those parts that
fit within the described process dynamics are useful to consider in the model (Figure 2). If
the distribution of sludge/water in systems with parallel lanes is well balanced, all streets 129
 I. Formulation of objectives

II. Process description

III. Data collection and

data verification
J. J.W. Hulsbeek et al.

IV. Model structure

V. Characterisation of flows

VI. Calibration

VII. Detailed characterisation

VIII. Validation

IX. Study

Figure 1 Main structure of the protocol

II. Process description

balanced model all lanes

Sludge-/water distribution N
division

Y
model one lane
Define process components

Overview in-/ outgoing flows

Expenditure: 1 - 2 days

Figure 2 Process description

can be modelled in the same way. If the distribution is not well balanced, each lane has to be
modelled separately. In most cases it is only required to describe the activated sludge
process (including secondary clarifier(s), but without e.g. the sludge handling facilities).
All in- and outgoing flows (i.e. influent, recirculation flows, internal flows from sludge
treatment processes and effluent) must be defined.

Data collection and data verification

In this phase the composition and the volume of the flows to the different process
components, as well as the volume of the process components are defined (Figure 3). It is
130 advised to gradually refine definitions, starting off with rough or approximating values and
 III. Data collection and data verification

Required data - volume of the involved process components;

- flows of all in- and outgoing streams;
- COD-concentration (at least influent and effluent);
- BOD-concentration (at least influent and effluent);
- phosphate-concentration (at least influent and effluent);
- nitrogen-concentration (N kj, NH 4-N en NO3-N, at least influent,
return sludge and effluent);

J. J.W. Hulsbeek et al.

- nitrogen- and phosphate concentration in the sludge;
- concentration suspended solids in the sludge and return sludge;
- concentration volatile suspended solids in the sludge;
- oxygen profile in the different process components (measurements
with mobile sensor);
- oxygen concentration in influent, return sludge and internal
recirculation if jack screws in use.

Data verification
mass balances
- suspended solids;
- COD;
- phosphorus;
- nitrogen.

Expenditure: 3 - 4 days

Figure 3 Data collection and data verification

fine-tune them in the process. In general, many data of the WWTP are available (daily
average concentrations, flow patterns). In the protocol it is advised, to initially generate the
compositions and the flows from the available data. If necessary, lacking relevant data can
then be obtained with extra monitoring. After a first set of simulations a better directed
monitoring programme can be developed. Concentrations which change most at specific
points in the treatment plant could be evaluated in detail whereas other values can be based
on daily (flow proportional) averages.
Using mass balances (Flows, COD, dry matter, nitrogen and phosphorus) the available
data should be checked (Meijer et al., 2001b). If the mass balances are not correct it is
necessary to perform extra monitoring of the plant. If these mass balances in the prime data
do not hold, calibration will by definition lead to erroneous results (Meijer et al., 2001a;
Nowak et al., 1999). Meijer et al. (2001b) present a detailed overview of this procedure
including the use of data reconciliation techniques. Mass balances for phosphorus (or iron
if chemical P-removal is applied) can be made simply by measuring influent and effluent
total phosphate concentrations, flows and by measuring the phosphate in the (excess)
sludge. Based on this mass balance for phosphorus the mass of excess sludge can be
checked or obtained if proper data are unavailable. Since the assumed SRT is highly
sensitive in the simulation model it is essential that this mass balance is scrutinised. Also if
one is not interested in modelling P-removal.
The (measured) flows can be checked and unknown flows can be calculated by using a
combination of the measurements of the concentration of suspended solids in the reactor or
incoming flows and the flows that are known. E.g. by:

Ga = ((Qrs * Grs) + (Qi * Gi))/(Qi + Qrs) (1)

with: Ga: suspended solids in a reactor (kg MLSS/ m3); Qrs: flow return sludge (m3/d); Grs:
suspended solids in return sludge (kg MLSS./m3); Gi: suspended solids in the influent of a 131
 reactor (kg MLSS/m3); and Qi: influent flow (m3/d). Depending on the plant configuration
more such balances can be applied (and checked).
For nitrogen the mass balance is more complicated,

Ni = Ne + Ns + Nd (2)

with Ni: total nitrogen load in the influent (kg/d); Ne: total nitrogen load in the effluent
J. J.W. Hulsbeek et al.

(kg/d); Ns: total nitrogen load in the excess sludge (kg/d); Nd: total denitrified nitrogen load
(kg/d).
For nitrification the following equation is valid:

Nki = Nke + Ns + Nn (3)

with Nki: total Kjeldahl-nitrogen load in the influent (kg/d), Nke: total Kjeldahl-nitrogen
load in the effluent (kg/d); Nn: total nitrified nitrogen load (kg/d). In practice Ni and Nki are
commonly the same.
The COD balance gives the oxygen uptake in the treatment plant:

CODi = CODe + OUR + (Nd * 2.86) + (Qs * Gs,org * 1.42) – (4.56 * Nn) (3)

with CODi: COD-load in the influent (kg O2/d); CODe: COD-load in the effluent (kg O2/d);
OUR: oxygen uptake rate (kg O2/d); 2.86 = oxygen reduction-equivalent (1 kg NO3-N is
equivalent with 2.86 kg O2); Gs,org: the concentration of volatile suspended solids in the
excess sludge (kg VSS/m3); Qs: the daily volume of excess sludge (m3/d); 1.42: the factor
to convert the amount of volatile suspended solids in COD equivalents (kg COD/ kgVSS);
4.56: conversion factor for the calculation of oxygen uptake during nitrification. The
obtained OUR can be used to check the supposed aeration efficiency of the plant
(kgO2/kWh) or to detect possible errors or use it in the model simulation set-up of the
aerators.

Model structure
The model structure for the hydraulics of the WWTP is based on the process description.
The definition of the model structure is an essential phase during the study. There is no
point calibrating if the model is not properly defined. Errors in e.g. flow rates of recircula-
tion pumps are more sensitive then errors in model parameters (Meijer et al., 2001a). In the
model definition the different process components are described. Aspects such as number
of compartments, aeration configuration, settling and control will be taken into account
during the definition of the model structure. In this phase it is essential to evaluate the
oxygen gradients in the aerated tanks in a horizontal and vertical direction. Certainly for
surface aeration this needs to be checked in order to set-up a proper compartmentation. For
bubble aeration a visual inspection of the operation of the aeration is advised. It is equally
important to control all flow splitters at the wwtp on exact operation. The modelling of the
settler depends heavily on the use of the model. The amount of suspended solids in
the effluent should be made equal to the measured values. Denitrification in the sludge line
can be checked by comparing the nitrate in the return sludge and effluent. This denitrifica-
tion can be best calibrated by applying a virtual tank in the return sludge line. Most settler
models do not calculate the biological processes, in general the denitrification in the settler
is more important then the exact height of the sludge blanket as a function of time. If a full
settler model is used an extra tank in the return sludge line for denitrification should not be
132 used since it would lead to a too high amount of total sludge in the total system. In general it
is required that the effects of the control are in accordance with the behaviour of the process
control in practice. One could e.g. define flow rates or DO based on the actually measured
values instead of implementing the controller. In Figure 4 the main parts of the model
structure are described.

Characterisation of the main flows

By using historical data and/or specific measurements, the important process flows can be

J. J.W. Hulsbeek et al.

characterised. These flows include influent, effluent, recirculation flows and internal
flows. The guidelines for influent characterisation are the basis for the characterisation
(Roeleveld, 2001) of the different flows. If the model is used for a system choice, daily
average concentrations of influent and effluent and the variations in the flow pattern are
sufficient. If the model is used for optimisation, the development of control strategies
specific data from 4 or 2 hour composite samples is required. In this case also other flows
than the influent and effluent, such as recirculation flows should be sampled. In Figure 5 the
procedure for the characterisation of the main flows is given.

Calibration
If the model definition and the minimal set of data are implemented in the activated sludge
model, a first calibration can be done. If the results of the calibration shows that a major
adjustment of the model parameters is necessary, experience has learned that it is most
likely that a structural error is present in the model. In that case it makes no sense to start a

IV. Model structure

Partition - partition of process component

- oxygen gradient

Aeration Calculation of the oxygen capacity with the COD

and nitrogen mass balances and comparing with
the practical values

Sedimentation Simple clarifier model

Prediction height of sludge-
(secondary clarifier ) layer necessary N SC-mix

Denitrification in
10 - layer model sec. clarifier N

Inclusion No
Development of control of a virtual denitrification
Process control N denitrification compartment
strategy
compartment
Y

.
Real process control Implementation of
programming process control to describe
the biol. processes right

Expenditure: 1-2 days

Figure 4 Model structure 133

 V. Characterisation of flows

Method According to Roeleveld 2001

Measuring duration - System choice study: 0 - 3 days

(influent characterisation ) - Optimisation study: 3 - 7 days
J. J.W. Hulsbeek et al.

- Development of a control strategy: 7 days

Measuring duration (internal flows) Depends on the influence of the various component on the
results

Type of samples - System choice study: 24 h samples

(influent characterisation) - Optimisation study: 2 - 4 h samples
- Development of a control strategy: 2 h samples

Expenditure: 1 - 2 days (data processing)

Figure 5 Procedure for the characterisation of the main flows

monitoring programme before the model structure is adjusted. In this case the balances
could be checked again and some extra verification measurements should be made.
If the effluent quality is initially not well predicted by the model, a sensitivity test for
several of the process parameters (flow rates, set-points etc.) can be made. This aids in
detecting where potential errors in the process have been made. By this method the number
of components that need to be analysed and the number of analyses can be reduced.
In the protocol it is advised to make a calibration according to the procedure given in
Figure 6. Further it is indicated what model parameters can be adapted to calibrate the
model properly:
• sludge composition and production: influent Xs and Xi, iNX, iNI;
• ammonium concentration in the effluent: KO2, kNH4, bA;
• nitrate concentration in the effluent: hN03, bH, KO2, KNO3, KOH.
First the N-content of the sludge in the simulation must be made equal to the measured
one. If not the error will accumulated in the nitrification/denitrification process. Since the
N-content of biomass is relatively well known, it is advised to adjust the N-content of
the inert particulates. This will affect also the definition of the influent composition. If the
SRT is checked and well known the sludge content in the treatment plant can be calibrated.
This is influenced most by the yield and decay rates and the assumed fraction of inert
particulates in he influent. It is advised to use the latter parameter as the only calibration
factor since it is anyway the most uncertain parameter.
The nitrification process is calibrated on the ammonium effluent content. After check-
ing that a wrong prediction of the ammonium is not caused by a wrong DO in the model or a
too low alkalinity, the affinity coefficients for oxygen or ammonium could be used to adjust
the ammonium in the effluent. For adjusting the ammonium in the effluent at different tem-
peratures it is advised to use the decay factor rather then the growth rate, because of the
larger uncertainty in the former parameter. One should pay attention not to calibrate the
model further then the accuracy of the actual measurements.
134 The denitrification process can be calibrated on the nitrate in the effluent. First it should
Conclusions
1. Based on the experience of a large and diverse group of practitioners in the Netherlands
a uniform protocol for the dynamic modelling of activated sludge systems has been set
up. This STOWA protocol is seen as a guideline for improving the quality and control-
lability of the simulation studies for activated sludge processes.
2. During the interviews with model-users bottlenecks were identified and analysed. For

VI. Calibration

Order of steps

1. Sludge production - sludge concentration

- control surplus
- sludge age
sludge flow
- sludge production
- XS /X I
k BZV
- N-content sludge - iNX iXI

2. Nitrification - O2 concentration - change oxygen cap.

- alkalinity - change alkalinity
in the influent
- NH 4 -concentration - K O2, K NH4, b A

- compare NO 3 -conc . in - implementation deni.

3. Denitrification
the effluent with the tank in return sludge
conc . in return sludge
- oxygenation screws - adjust partipition
and aeration system
- NO 3 -conc . in effluent - ηNO3 , bH, K O2,K NO3 , K OH

4. Internal flows - NH 4 -concentration - K O2 , K NH4

- NO 3 -concentration - η NO3

If during the calibration large adjustments from (kinetic) parameters are necessary (values out of the
range) then in most cases the system boundary is not well enough defined. In that case back to II.
of the protocol (process description) or back to III. (data collection and data verification).

Expenditure: 5 - 6 days

Figure 6 Order of steps during the calibration 135

 Table 1 The range of the various values and the default values of different model parameters for ASM 1

Symbol Description Default Range Unit

value

YH Heterotrophic yield 0.67 0.46–0.69 g COD/g COD

YA Autotrophic yield 0.24 0.07–0.28 g COD/g N
iNX Mass of N per mass of COD in biomass 0.086 g N/g COD
iNI Mass of N per mass of COD in inerts 0.06 0.02–0.1 g N/g COD
J. J.W. Hulsbeek et al.

µH Maximum specific growth rate for heterotrophic

biomass 6 3.0–13.2 l/d
KS Half saturation coefficient for heterotrophic biomass 20 10–180 g COD/m3
KOH Saturation/ inhibition coefficient for oxygen, for
heterotrophic biomass 0.2 0.01–0.20 g O2/m3
KNO Saturation/ inhibition coefficient for nitrate 0.5 g NO3-N/m3
bH Decay rate for heterotrophic biomass 0.62 0.05–1.6 l/d
bA Decay rate for autotrophic biomass 0.15 l/d
hg Reduction factor for anoxic growth 0.8 –
hh, hNO3 Reduction factor anoxic hydrolyses 0.4 0.6–1.0 –
kH Maximum specific hydrolysis rate 3.0 1.0–3.0 l/d
KX Saturation coefficient for particulate COD 0.03 0.01–0.03 g COD/g COD
µ Maximum specific growth rate for autotrophic
biomass 0.8 0.34–0.8 l/d
KNH4 Saturation coefficient for ammonium 1.0 g NH4-N/m3
KOA Saturation coefficient for oxygen for autotrophic biomass 0.4 g O2/m3

References
Meijer, S.C.F., Van Loosdrecht, M.C.M. and Heijnen, J.J. (2001a). Metabolic modelling of full scale biolog-
ical nitrogen and phosphorus removing wwtp. Water Research 35, 2711–2723.
Meijer, S.C.F., Van der Spoel H., Heijnen, J.J. and van Loosdrecht M.C.M. (2001b). Error diagnostics and
data reconciliation for Activated sludge modelling purposes using linear conservation relations. Wat.
Sci. Tech. This issue.
Nowak, O., Franz, A., Svardal, K., Muller, V. and Kuhn, V. (1999). Parameter estimation for activated
sludge models with the help of mass balances. Wat. Sci. Tech. 39(4), 113–120.
Roeleveld, P. and Kruit, J. (1998). Richtlinien für die Charakterisierung von Abwasser in den Niederlanden,
Korrespondenz Abwasser, 3, 465–468, in German.
Roeleveld, P.J. and van Loosdrecht M.C.M. (submitted). Experiences with Guidelines for Wastewater
Characterisation in the Netherlands. Wat. Sci. Tech. This issue.
STOWA (1996). Methods for influent characterisation, inventory and guidelines, report nr. 96–08, in Dutch.
STOWA (1999). Influent characterisation of raw and pre-treated wastewater, report nr. 99–13, in Dutch.

136