594 © IWA Publishing 2013 Water Science & Technology | 67.3 | 2013

Computer simulation of control strategies for optimal

anaerobic digestion

S. Strömberg, M. O. Possfelt and J. Liu

ABSTRACT

Three previously published control strategies for anaerobic digestion were implemented in Simulink/

Matlab using Anaerobic Digestion Model No. 1 (ADM1) to model the biological process. The

controllers’ performance were then simulated and evaluated based on their responses from five

different types of process scenarios i.e. start-up and steady state performance as well as

disturbances from concentration, pH and ammonia in the inflow. Of the three evaluated control

strategies, the extremum-seeking variable gain controller gave the best overall performance.

However, a proportional feedback controller based on the pH-level, used as a reference case in the

evaluation, proved to give as good results as the extremum-seeking variable gain controller but with

a lower wear on the pump. It was therefore concluded that a fast proportional control of the reactor

pH is a key element for optimally controlling a low-buffering anaerobic digestion process.

doi: 10.2166/wst.2012.603

S. Strömberg (corresponding author)M. O. PossfeltJ. LiuDepartment of Biotechnology,Lund University,Getingevägen 60,221 00 Lund,SwedenE-mail: sten.stromberg@biotek.lu.se

J. LiuBioprocess Control Sweden AB,Scheelevägen 22,223 63 Lund,Sweden

Keywords | anaerobic digestion (AD) control, ADM1, anaerobic digestion, biogas, computer simulation

NOMENCLATURE

Extremum-seeking variable gain controller

Q

Inflow

Q0

Inflow at set point

KpH

Proportional gain for pH

KCF

Proportional gain for gas flow

pH

pH level

pHsp

pH set point

pHsp,o

pH set point at set point

GF

Total gas flow

GFsp

Total gas flow set point

GFsp,old

Previous total gas flow set point

GFstep

Change in gas flow set point

D

Difference between GF and GFsp Dmin Lower boundary for D

Dmax

Upper boundary for D

Dold

Previous value of D

Disturbance monitoring controller

ΔQ

Change in inflow

Qold

Previous inflow

EGV

Gas volume from pulse

EGVexp

Expected EGV

α

Increase in flow rate in pulse

δt

Time for pulse

GFav

Average gas flow

R

Ratio of EGV and EGVexp

Rmin

Lower boundary for R

Rmax

Upper boundary for R

fmax

Maximal change in inflow

Hydrogen variable gain controller

ΔD

Change in dilution rate

Dold

Previous dilution rate

k

Proportional gain

fCH4

Factor based on CH4 productivity

fH2

Factor based on H2 concentration

α

Sensivity parameter for fCH4

GFCH4

Methane productivity

GFCH4,sp

Methane productivity set point

H2

Hydrogen concentration in gas

H2,sp

Hydrogen concentration set point

n

Sensivity parameter for fH2at high H2

m

Sensivity parameter for fH2at low H2

595 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

INTRODUCTION

Anaerobic digestion is a complex and versatile biologicalprocess that needs constant monitoring in order to achieveits optimal performance. In case of a digester failure, the

process could take up to several months to recover, leadingto big losses in revenues (Bernard et al. ). Due to this, tobe on the safe side, many operators prefer to operate the

biogas plant far below the maximum capacity. A way toovercome this insecurity whilst still maintaining a stable pro-cess with a higher gas productivity is to implement a control

strategy that can push the process to an optimum and at thesame time have the capacity to react fast on disturbancesand unstable process conditions (Steyer et al. ; Liuet al. ; Rodríguez et al. ).

Over the past decades, several control strategies havebeen developed for the control of anaerobic digestion. Inmany of these cases, the controller was able to improve

the process significantly, either by maximizing the gas pro-duction and/or by disturbance rejection. However, so far,little focus has been directed toward comparing and evaluat-

ing these control strategies, much due to the extensiveworkload of such studies. A way to overcome this problemis to utilize dynamic computer simulations (Rosen et al.). Model implementations using Anaerobic DigestionModel No. 1 (ADM1) has proved to be a good tool for eval-uating control performance in an anaerobic digestionprocess (Batstone & Steyer ; Zhou et al. ). As an

example, the Benchmark Simulation Model No. 1 (BSM1,Copp et al. ), a benchmark platform for evaluation ofcontrol strategies and operation of wastewater treatment

plants, has been developed for this purpose and has hadgreat success with over 300 publications worldwide(Nopens et al. ). With inspiration from BSM1, this

study aimed at evaluating three previously publishedcontrol strategies by putting them through a number ofcritical process situations with the help of computersimulations.

Table 1 | Summary of evaluated controllers

Controller Reference

Open loop reference controller �Proportional reference controller �Extremum-seeking controller Liu et al. (,

Disturbance monitoring controller Steyer et al. ()

Hydrogen-based variable gain controller Rodríguez et al. (

METHODS

Evaluated control strategies

In order to find the most suitable controllers to evaluate, aliterature study was carried out. It was required that the con-trollers would be unique, reject disturbances and maximize

the biogas or methane production. Especially the maximiz-ing criteria greatly limited the number of candidates, andin the end three strategies where chosen (Table 1).

Below is a short summary of the control algorithms eval-uated in this work. For a more detailed description andexplanation of the symbols please see the respective refer-

ence (Table 1).

Extremum-seeking variable gain controller (ES)

This control strategy uses a cascade structure with anadditional rule-based expert system where the lower levelcontroller (also called the slave controller) operates on the

pH-level and the top level controller (also called themaster controller) manipulates the set point of total gasflow while the expert system pushes the gas production.

Q ¼ KpH � pH � pHsp� �þQo (1)

pHsp ¼ KGF � GF �GFsp� �þ pHsp,D (2)

If D � Dmax then GFsp ¼ GFsp,old þGFstep (3)

If Dmin � D<Dmax then GFsp ¼ GFsp,old þ 0:5GFstep

(4)

If D<Dmin & Dold � Dmax then GFsp ¼ GFsp,old (5)

If D<Dmin & Dold <Dmax then GFsp ¼ GFsp,old �GFstep

(6)

Input Output Acronym

� Flow rate OR

pH Flow rate PR

) pH & gas prod. Flow rate ES

pH & gas prod. Flow rate DM

) H2 & CH4 prod. Flow rate HVG

596 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

The lower level controller regulates the inflow accord-

ing to Equation (1), in which the gain constant (KpH) isdetermined by a gain scheduling state function with eightdifferent process states (Liu et al. ). The objective of

the top level controller is to make sure that the gas flow,set by the supervisory logic from the expert system, is main-tained. This objective is achieved by manipulating the setpoint of the lower level controller according to Equation

(2). The supervisory logic is achieved by a simple set ofrules that can be seen in Equations (3)–(6), where Dmax

and Dmin are constants and D is the gas flow (GF) sub-

tracted by the gas flow set point (GFsp). In Equations(1)–(6) the subscript old refers to the previous value ofthe parameter and step refers to the change that is added

to the parameter. In the numerical implementation the fol-lowing sample times were used: 2.5 min for lower levelcontroller, 30 min for upper level controller and 60 minfor the rule-based expert system.

Disturbance monitoring controller (DM)

The basic concept of the disturbance monitoring controller

is to introduce a disturbance on purpose to the process andthen, based on the response, determine if the process isoperating at maximum capacity, or is above/below maxi-

mum (Velut et al. ). The disturbance is applied as ashort pulse of increased inflow and the change in gas pro-duction is used as the response variable. The size of thepulse is calculated as a factor (α) multiplied with the cur-

rent inflow and duration of the pulse (δt). Based on theresponse from the pulse, the change in inflow is adjustedaccording to a number of equations (Equations (7)–(11)).

At first, the expected gas production increase (EGVexp) iscalculated by taking into account the pulse size and theaverage gas production just before the pulse (Equation

(7)). In order to determine the state of the process, a ratioof the expected and the actual gas production is calculated(Equation (8)) which is then used to determine the changein inflow according to Equations (9)–(11). The controller

also has an emergency pH stop which prevents furtherpulses if the pH falls below a critical level. In the numericalimplementation, a 1 hour pulse was generated at an inter-

val of 0.4 days.

EGVexp ¼ α � δt �GFav (7)

R ¼ EGVEGVexp

(8)

If R � Rmax then ΔQ ¼ Qold � fmax � R� Rmax

1� Rmax

� �(9)

If Rmin � R< Rmax then ΔQ ¼ 0 (10)

If R< Rmin then ΔQ ¼ Qold � fmax � R� Rmin

Rmin

� �(11)

Hydrogen-based variable gain controller (HVG)

This control strategy uses hydrogen concentration fromthe reactor headspace and methane productivity to manip-

ulate the inlet flow rate. The change in dilution rate(Equation (12)) is a function of the old dilution rate(Dold), a gain constant (K) and two empirical functions;one based on the methane productivity ( fCH4

, Equation

(13)) and one based on the hydrogen concentration ( fH2,

Equations (14)–(15)).

ΔD ¼ Dold �K � fH2 � fCH4 � Δt (12)

fCH4 ¼α �GFCH4,sp

QCH4 þ α �GFCH4,sp∈ 0, 1½ � (13)

If H2 >H2,sp then fH2 ¼H2,sp

H2

� �n

�1 ∈ �1, 0½ � (14)

If H2 � H2,sp then fH2 ¼ 1� H2

H2,sp

� �1m ∈ 1, 0½ � (15)

The main objective of fCH4is to determine the size of the

change in inflow whereas fH2, besides determining the size

of the inflow, also decides the direction of the changein inflow (increase or decrease). The controller was

implemented in a pseudo-continuous mode, i.e. Δt inEquation (12) was varied according to the time step in thenumerical solver.

Evaluated process situations

The controllers were evaluated based on their responses tofive different types of process scenarios. These werechosen to simulate known critical situations that can

occur in the operation of an anaerobic reactor and are allpresented below.

Table 2 | The run protocols used in the simulations

Time span(d) Purpose Disturbances

Start-up and concentration disturbances

0–100 Time for start-up as well as for reachingsteady state

No

100–282 Time for the controller to adapt to thedisturbances

Yes

597 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

Start-up

Evaluates how fast and stable the controllers can accom-plish a start-up of an anaerobic process. This was

interesting to study since the start-up of a reactor is a verytime consuming and sensitive process that differs from thesteady state conditions of an ongoing process (Zhao et al.).

282–646 Evaluation period Yes

Steady state performance and pH and ammonia disturbances

0–63 Time for reaching steady state No

63–427 Evaluation period Yes

Steady state performance

Evaluates how far the controllers can push the process atsteady state when the inlet concentration is constant.

Concentration disturbances

Evaluates how well the controllers can handle big changesin the inlet concentrations and was implemented as a

random gain on all the inlet concentration states, i.e. chan-ging every second day within a span of 0.1 to 2.

pH disturbances

Evaluates how well the controllers can handle big changesin the inlet pH. Two types of disturbance were tested: (1)pulses of low pH that grew bigger with time to see the con-

trollers’ response to quick changes in pH and (2) a negativepH gradient to see how far the controller can sustain a work-ing process.

Table 3 | Evaluation parameters and their descriptions

Evaluationparameter Description

Ammonia disturbances

Evaluates how well the controllers can handle big changesin the inlet ammonia concentration. This was interestingto study due to the different inhibition properties of ammo-

nia, where inhibition occurs at high pH levels and mainly onthe methanogenic bacteria. The ammonia disturbances wereadded in the same way as the pH disturbances, i.e. as grow-

ing pulses and as a positive gradient.

tstart–up Time for the process to reach steady state from

the initial inoculum

QCH4Average methane productivity during theevaluation period

CODred Average chemical oxygen demand (COD)reduction during the evaluation period

A.w. Actuator wear of the pump

Run protocol

In order to evaluate the performance of the control strategies

as objectively as possible a run protocol, specifying differentphases in the simulation runs, was designed (Table 2).

Evaluation criteria

The control systems were evaluated based on the resultsfrom four process parameters which are presented anddescribed in Table 3.

The actuator wear was calculated according to Equation(16) where tstart and tstop are the start and stop time for the

evaluation, @Q@t

��� ��� is the derivative of the inflow (controlleroutput) and n is the number of simulation time stepswithin the evaluation period.

A:W : ¼Ptstop

tstart

@Q@t

��������

n(16)

Anaerobic digestion model

ADM1 (Batstone et al. ) was used to model the bio-logical and physio-chemical processes in the reactor.

This model has been widely used and on several

598 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

occasions proved to provide an accurate representation of

the process (Parker ; Batstone et al. ). However,in order to achieve a more realistic start-up process, vari-able growth rates of the bacterial groups were added to

the model. These were implemented as a time-varyinggrowth rate factor that was multiplied with the growthrate (Zhao et al. ). In order to make it as simple aspossible, the same growth rate factor was used for all bac-

terial groups and was designed as an exponential curvegoing from zero to one during the first 30 days.

Equalization tank

An equalization tank previous to the digester was included

in order to give the controller a realistic amount of substrateto work with. It was implemented as a simple mixing modelwhere the derivatives were a function of the inputs and out-puts of each component.

Measurement noise and filters

To increase the realism of the simulated process and toavoid giving different control strategies unrealistic advan-tages, measurement noise and filtering were added to the

input signals. Measurement noise was added as a 2.5%random error of the original signal (Rieger et al. ). Thesignals were filtered using a backward discretized low-pass

filter with a step size of 0.1 min and a filter time constantof 10 min. Also, a negative drift of the pH sensors wasincluded as a gradual decrease of 0.1 pH unit/month com-pared to the real pH-value. This drift phenomena was then

reset every second month when the sensor was assumedto be recalibrated.

Substrate

The substrate characteristics were the same as presented in

(Rosen & Jeppsson ), with a low buffering capacity andthe main biodegradable material in the form of carbo-hydrates, proteins and lipids.

Controller tuning

The tuning of the control parameters was primarily focused

on maximizing the methane production in the simulationswhere concentration disturbances were applied. In theprocedure, a high average methane productivity and disturb-

ance rejection capability were prioritized compared to amore stable and low producing process. Since all evaluated

control algorithms were structured differently, no uniform

tuning method could be derived. Instead, a large numberof simulations were implemented for familiarization withthe controller performances and, when sufficient experience

had been gained, the parameters were tuned manually tofulfil the requirements mentioned above. This simplemethod was chosen to reduce the otherwise extensive com-plexity of the calibration procedure. If the controllers are to

be evaluated for a more specific target process, the cali-bration should be matched to the conditions of thatprocess in order to reach the optimal performance of the

controllers.

Numerical implementation

The simulations were carried out in Simulink/Matlab usingode45 as the numerical solver. ADM1 was implemented inC-code of the DAE2 version (s-functions were provided by

Dr Ulf Jeppsson, Department of Industrial Electrical Engin-eering and Automation, Lund University) for maximalsimulation speed (Rosen & Jeppsson ).

RESULTS

Below is a short summary of the results from the simulations

of the five different presented process scenarios. Due to alack of space, most of the visualization of the results hadto be excluded from this paper.

Start-up

Only ES and PR (proportional reference controller) were

able to complete the start-up within the designated timeframe (Table 4). These results indicate that a simple pro-portional controller should be able to handle a start-up

process fairly well. The large values for DM and HVG indi-cate that these controllers are too slow to fully optimize astart-up process.

Steady state performance

In the steady state performance evaluation ES and DM wereable to give fairly good results compared with the open loop

reference, OR (Table 4). All the controllers had a typicaltrend of a decrease in COD reduction with an increase ingas productivity and vice versa. PR, ES and DM gave similar

results in methane productivity and COD reduction whereasthe actuator wear differed significantly.

Table 4 | Summary of most important results from simulations of start-up, steady state and concentration disturbances

Start-up time (days) Average QCH4(Nm3/m3/day) Average CODred (%) Average A.w. (-)

Controller Conc. constant Conc. constant Conc. disturbance Conc. constant Conc. disturbance Conc. constant Conc. disturbance

OR � 1.29 1.31 38.4 12.2 0 0

PR 64 1.44 1.46 33.3 30.6 1,770 2,020

ES 60 1.43 1.44 34.6 31.1 7,090 7,880

DM 100 1.45 0.66 33.4 34 793 270

HVG 100 1.26 1.16 38.9 39.8 1.09 76.9

599 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

Concentration disturbances

In the concentration disturbance evaluation, ES had thehighest methane productivity (1.44 Nm3/m3/d) of the evalu-ated controllers (Table 4). One interesting point discovered

was that the methane productivity for ES was at the samelevel as in the steady state performance evaluation, indicat-ing that it can handle varying inlet concentration very

well. In contrast, DM and HVG both had lower methaneproductivities compared to the open-loop case, demonstrat-ing that these controllers experienced difficulties with thedisturbances.

As with the steady state case, ES experienced higheractuator wear compared to the other controllers and thiscan be clearly seen when comparing the volumetric inflow

from ES (c) and PR (d) in Figure 1. A more oscillatingtrend from ES can also be seen when comparing themethane productivity (e and f ) while the two controllers

had a more similar profile with regards to the methanogenicbiomass concentration (sum of acetogenic and hydrogeno-trophic groups in ADM1) (g and h). A comparison of the

pH-level trend (i and j) indicates that PR maintained aslightly lower pH level but with a similar trend as ES.

pH disturbances

The results from the pH disturbances were, as expected,much worse with regard to both methane productivity and

COD reduction compared to the results from the concen-tration disturbances (Table 5). The only evaluatedcontroller that had a reasonable methane productivity wasES. This was especially obvious in the simulations with

pulse disturbances. Once again, PR gave a similar methaneproductivity and COD reduction as ES. Another pointworth mentioning is that HVG had a rather high COD

reduction even though the methane productivity was verylow.

Ammonia disturbances

Similar to the pH disturbances, the results from the ammo-nia disturbances also show low methane productivities forall controllers (Table 5). The difference was that the results

were more evenly distributed among the controllers, but inmany of the previous cases, ES had the highest methaneproductivity. The controller that deviated the most was

HVG which had a lower methane productivity comparedto OR.

DISCUSSION

Since this work has been carried out strictly in silico, with-

out any form of experimental data, there are severallimitations and uncertainties that should be consideredwhen evaluating the results. Three of the most importantones are: (1) even though ADM1 is a very advanced

model, it is still far from describing the full complexity andunpredictability of a real-life anaerobic digestion process;(2) the same type of substrate characteristics is used in all

simulations, giving very uniform responses; (3) the tuningof the control parameters was performed manually, with arather simple methodology, which compromises the validity

when comparing the control strategies. Therefore, to betterverify the results, an improved strategy for tuning the controlparameters needs to be designed and implemented. How-

ever, this should be considered a difficult task as mostcontrollers for anaerobic digestion are non-linear anddiffer greatly in both structure and type.

ES was found to give the best overall performance of

the evaluated control strategies. However, ES alsohad the highest actuator wear. Interestingly, PR gavethe best results of all controllers but, as ES, experienced

high values for the actuator wear. The similar resultsfrom ES and PR are probably due to the fact that

Figure 1 | Examples of some results from the simulations of concentration disturbances with inlet COD concentrations for ES (a) and PR (b), inflow for ES (c) and PR (d), methane pro-

ductivity for ES (e) and PR (f), methanogenic biomass for ES (g) and PR (h) and pH for ES (i) and PR (j).

600 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

Table 5 | Summary of the most important results from the simulations of pH and ammonia disturbances

Average QCH4(Nm3/m3/day) Average CODred (%) Average A.w. (-)

ControllerpHpulse

pHgrad.

NH3

pulseNH3

grad.pHpulse

pHgrad.

NH3

pulseNH3

grad. pH pulse pH grad. NH3 pulse NH3 grad.

OR 0.04 0.09 0.41 0.57 1.83 3.23 12.8 17.5 0 0 0 0

PR 0.94 0.44 0.51 0.64 35.4 48.2 9.48 16.1 7.42×107 7.4×107 7.42×107 7.42×107

ES 0.92 0.45 0.52 0.63 37.5 48.2 9.76 16.8 5.76×108 5.76×108 5.76×108 5.76×108

DM 0.04 0.02 0.47 0.45 8.64 10.7 9.63 9.73 23.4 0.703 598 –

HVG 0.04 0.34 0.28 0.46 39.9 40.1 8.62 20.0 4.03 17 4,410 3.51

601 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

both have a fast feedback loop acting on the pH-level. The

high actuator wears for both these controllers show thatthey were affected by the measurement noise and/orexperienced instability around the set point which mostlikely could be reduced by using longer sampling inter-

vals, better filters and differently tuned parameters.Nevertheless, the good performance from these two con-trollers indicates that a reasonable parameter for both

disturbance rejection and gas maximization is the pH-level, provided that on-line pH monitoring can be prop-erly implemented. However, this conclusion is limited to

the specific feedstock and conditions used in this study.pH as a process indicator has on many occasions beenproven to be an unreliable parameter due to different buf-

fering capacities of the feedstock and if a high bufferingfeedstock would have been investigated, the outcomemight have been different (Strik et al. ). Therefore,to evaluate the controllers’ performance on a more gen-

eral level, several similar studies, using different types offeedstock, should be carried out.

Both the disturbance monitoring controller (DM) and

the hydrogen-based variable gain controller (HVG) hadproblems handling the different disturbance scenariosand also experienced long start-up times. One of the

major reasons for this was likely a poor tuning of the con-trol parameters. HVG in particular was complicated totune since its parallel controller structure made it difficultto foresee the sensitivity and effects of each parameter. For

DM, the much longer sampling interval (only 2.5 pulsesper day) was probably one of the major reasons for thepoor results. Another concern with DM is that the probing

pulses themselves might lead to reactor imbalance. A wayto reduce this risk could be to include negative pulses(Velut et al. ).

An aspect that was not fully covered in this study is thestability. Since all controllers except PR have variableset points it was difficult to evaluate this in detail.

Nevertheless, when they were tested with step disturbances

almost no fluctuations were observed even though someovershooting occurred. The controllers were also able tohandle measurement noise without failing even if this intro-duced fluctuations in the controller output. The actuator

wear gives an indication of the controller’s sensitivity tonoise and other disturbances, and controllers with highactuator wears should therefore be used with caution if

implemented in a real-case application. However, when itcomes to the stability of the anaerobic digestion (AD) pro-cess, the controllers that experienced high actuator wears

(i.e. ES and PR) were also able to reject disturbancesbetter and thus maintain a more stable process.

A problem that needs to be further addressed is how to

evaluate the performance of a control system. An importantfactor is the objective of the process, that is, whether it is tomaximize the gas production or reduce a waste stream asmuch as possible. On many occasions these two objectives

contradict each other. When the process is pushed to gener-ate as much gas as possible, less material will have thechance to be fully degraded and accumulation of VFA

(volatile fatty acids) will occur (Ahring et al. ). Thiswill in turn leave high amounts of easy degradable materialin the effluent which is undesired if the feedstock is to be

degraded as much as possible. Therefore, if the aim is maxi-mal degradation, pH could be less interesting to use, sincethis parameter has been shown to give small changes instable and less stressed processes (Björnsson et al. ;

Boe et al. ).

CONCLUSIONS

A fast control strategy based on pH was found to give thebest overall performance in an evaluation based on simu-

lations using ADM1. In the simulations, the controlsystems were tested with severe disturbances in the

602 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

incoming feedstock. The conclusion was that, besides the

good process indications from pH, a short sampling timeof the control loop was especially important to handle thedisturbances. Another point that was found to be important

for the performance is how the control strategy is structured.In fact, in theory most control systems should be able tohandle disturbances rather well. However, in order to dothis, every control parameter has to be carefully tuned

which requires expert knowledge and understanding of thecontrol mechanisms. Such knowledge and understandingis much easier to gain when the structure of the controller

is simple and clear. This fact had a significant impact onthe outcome of this study, where the most successful controlstrategies had a more well-defined structure which made

their tuning more efficient.Using computer simulations to evaluate control strat-

egies for the control of anaerobic digestion has proved tobe a promising approach as long as the user is aware of

the limitations of such studies. However, to reach its fullpotential, the simulations platform needs to be developedfurther to include more features such as new types of feed-

stock, generic performance indicators and a better definedtuning strategy.

ACKNOWLEDGEMENTS

The authors would like to thank the Swedish Energy Agency

(2008–000449) for funding this project.

REFERENCES

Ahring, B. K., Sandberg, M. & Angelidaki, I. Volatile fattyacids as indicators of process imbalance in anaerobicdigestors. Applied Microbiology and Biotechnology 45,559–565.

Batstone, D. J. & Steyer, J. P. Use of modelling to evaluatebest control practice for winery-type wastewaters. WaterScience and Technology 56 (2), 147–152.

Batstone, D. J., Keller, J., Angelidaki, I., Kalyuzhnyi, S. V.,Pavlostathis, S. G., Rozzi, A., Sanders, W. T. M., Siegrist, H.& Vavilin, V. A. Anaerobic Digestion Model No. 1(ADM1), IWA Task group for Mathematical Modelling ofAnaerobic Digestion Processes. IWA Publishing, London.

Batstone, D. J., Keller, J. & Steyer, J. P. A review of ADM1extensions, applications and analysis: 2002–2005. WaterScience and Technology 54 (4), 1–10.

Bernard, O., Chachuat, B., Hélias, A., Le Dantec, B., Sialve, B.,Steyer, J. P., Lardon, L., Neveu, P., Lambert, S., Gallop, J.,Dixon, M., Ratini, P., Quintabà, A., Frattesi, S., Lema, J. M.,

Roca, E., Ruiz, G., Rodriguez, J., Franco, A., Vanrolleghem,P., Zaher, U., De Pauw, D. J. W., De Neve, K., Lievens, K.,Dochain, D., Schoefs, O., Fibrianto, H., Farina, R., AlcarazGonzalez, V., Gonzalez Alvarez, V., Lemaire, P., Martinez,J. A., Esandi, F., Duclaud, O. & Lavigne, J. F. Anintegrated system to remote monitor and control anaerobicwastewater treatment plants through the internet. WaterScience and Technology 52 (1–2), 457–464.

Björnsson, L., Murto, M. & Mattiasson, B. Evaluation ofparameters for monitoring an anaerobic co-digestion process.Applied Microbiology and Biotechnology 54, 844–849.

Boe, K., Batstone, D. J., Steyer, J. P. & Angelidaki, I. Stateindicators for monitoring the anaerobic digestion process.Water Research 44, 5973–5980.

Copp, J. B., Alex, J., Béteau, J. F., Dudley, J., Dupont, R., Gillot, S.,Jeppsson, U., LeLann, J. M., Pons, M. N. & Vanrolleghem,P. A. The COST Simulation Benchmark: Descriptionand Simulator Manual. Office for Official Publications of theEuropean Communities, Luxembourg.

Liu, J., Olsson, G. & Mattiasson, B. Monitoring and controlof an anaerobic upflow fixed-bed reactor for high-loading-rateoperation and rejection of disturbances. Biotechnology andBioengineering 87 (1), 43–53.

Liu, J., Olsson, G. & Mattiasson, B. Extremum-seeking withvariable gain control for intensifying biogas production inanaerobic fermentation. Water Science and Technology 53(4–5), 35–44.

Nopens, I., Benedetti, L., Jeppsson, U., Pons, M. N., Alex, J., Copp,J. B., Garnaey, K. V., Rosen, C., Steyer, J. P. & Vanrolleghem,P. A. Benchmark simulation model no 2: Finalisation ofplant layout and default control strategy. Water Science andTechnology 62 (9), 1967–1974.

Parker, W. J. Application of the ADM1 model to advanceanaerobic digestion. Bioresource Technology 96 (16), 1832–1842.

Rieger, L., Alex, J., Winkler, S., Boehler, M., Thomann, M. &Siegrist, H. Progress in sensor technology – progressin process control? Part I: Sensor property investigationand classification. Water Science and Tecnology 47 (2),103–112.

Rodríguez, J., Ruiz, G., Molina, F., Roca, E. & Lena, J. M. Ahydrogen-based variable gain controller for anaerobic digestionprocesses. Water Scicence and Technology 54 (2), 57–62.

Rosen, C. & Jeppsson, U. Aspects on ADM1 implementationwithin the BSM2 framework. Technical Report CODEN:LUTEDX/(TEIE-7224)/1-35/(2006), Dept of IndustrialElectrical Engineering and Automation, Lund Institute ofTechnology.

Rosen, C., Jeppsson, U. & Vanrolleghem, P. A. Towards acommon benchmark for long-term process control andmonitoring performance evaluation. Water Science andTechnology 50 (11), 41–49.

Steyer, J. P., Buffière, P., Rolland, D. & Moletta, R. Advancedcontrol of anaerobic digestion processes through disturbancemonitoring. Water Research 33 (9), 2059–2068.

Strik, D. P. B. T. B., Domnanovich, A. M. & Holubar, P. ApH-based control of ammonia in biogas during anaerobic

603 S. Strömberg et al. | Computer simulation of control strategies Water Science & Technology | 67.3 | 2013

digestion of artificial pig manure and maize silage. ProcessBiochemistry 41, 1235–1238.

Velut, S., de Maré, L. & Hagander, P. Bioreactor controlusing probing feeding strategy and mid-ranging control.Control Engineering Practice 15 (2), 135–147.

Zhao, B. H., Mu, Y., Dong, F., Ni, B. J., Zhao, J. B., Sheng, G. P.,Yu, H. Q., Li, Y. Y. & Harada, H. Dynamic modeling the

anaerobic reactor startup process. Industrial and EngineeringChemistry Research 49, 7193–7200.

Zhou, H., Li, H. & Wang, F. Anaerobic digestion of differentorganic wastes for biogas production and its operationalcontrol performed by the modified ADM1. Journal ofEnvironmental Science and Health, Part A: Toxic/HazardousSubstances and Environmental Engineering 47 (1), 84–92.

First received 3 April 2012; accepted in revised form 10 September 2012