Anda di halaman 1dari 13

Bioresource Technology 174 (2014) 243–255

Contents lists available at ScienceDirect

Bioresource Technology
journal homepage: www.elsevier.com/locate/biortech

Performance optimization and validation of ADM1 simulations under


anaerobic thermophilic conditions
Nabil M. Atallah a, Mutasem El-Fadel a,⇑, Sophia Ghanimeh b, Pascal Saikaly c, Majdi Abou-Najm a
a
Department of Civil and Environmental Engineering, American University of Beirut, Lebanon
b
Department of Civil and Environmental Engineering, Notre Dame University – Louaize, Lebanon
c
Water Desalination and Reuse Center and Division of Biological and Environmental Sciences and Engineering, King Abdullah University of Science and Technology, Thuwal
23955-6900, Saudi Arabia

h i g h l i g h t s

 ADM1 simulations of thermophilic anaerobic digesters treating food waste.


 Intermediary output influenced by different parameters depending on related processes.
 Methane-based calibration is less accurate in simulating intermediary by-products.
 Multiobjective optimization provided better overall results than methane optimization.
 Optimization results were validated upon their application on independent experiments.

a r t i c l e i n f o a b s t r a c t

Article history: In this study, two experimental sets of data each involving two thermophilic anaerobic digesters treating
Received 8 August 2014 food waste, were simulated using the Anaerobic Digestion Model No. 1 (ADM1). A sensitivity analysis was
Received in revised form 26 September conducted, using both data sets of one digester, for parameter optimization based on five measured per-
2014
formance indicators: methane generation, pH, acetate, total COD, ammonia, and an equally weighted
Accepted 29 September 2014
Available online 13 October 2014
combination of the five indicators. The simulation results revealed that while optimization with respect
to methane alone, a commonly adopted approach, succeeded in simulating methane experimental
results, it predicted other intermediary outputs less accurately. On the other hand, the multi-objective
Keywords:
Thermophilic anaerobic digestion
optimization has the advantage of providing better results than methane optimization despite not cap-
Food waste turing the intermediary output. The results from the parameter optimization were validated upon their
ADM1 independent application on the data sets of the second digester.
Ó 2014 Elsevier Ltd. All rights reserved.

1. Introduction Ward et al., 2008; Jeong et al., 2005). This problem is more pro-
nounced in thermophilic systems where the high temperature
Anaerobic digestion (AD) is a biological treatment process that (50–55 °C) increases degradation rates and speeds up the release
involves a series of synergetic biochemical pathways that degrade of VFAs. As a result, long-term operation of thermophilic digesters,
organic matter into a methane-rich gas that can be used as a source fed only with food waste, is often subject to instability and, in some
of energy (Cesur, 2004). While it was initially applied in the treat- instances, irreversible accumulation of inhibiting metabolic
ment of liquid wastes with low solids content, its application byproducts. In this context, the prediction of such disturbances
evolved to targeting higher solid content wastes, such as food and reduction of their occurrences become critical to a successful
waste (El Fadel et al., 2012) albeit the drawback of system instabil- digester operation. For this purpose, the Anaerobic Digestion
ity associated with the fast release of Volatile Fatty Acids (VFA) and Model (ADM) series were developed by the International Water
high ammonia levels (Ghanimeh et al., 2013; Banks et al., 2008; Association to test various biochemical reactions in the AD process
including disintegration, hydrolysis, substrate uptake and decay, as
well as physiochemical processes such as association/dissociation
⇑ Corresponding author at: American University of Beirut, Faculty of Engineering
and liquid–gas transfer (Esposito et al., 2011). ADM1 has report-
and Architecture, Bliss Street, PO Box 11-0236, Beirut, Lebanon. Fax: +961 1 744
462. edly been successfully used in simulating AD of various types of
E-mail address: mfadel@aub.edu.lb (M. El-Fadel). waste, including, but not limited to, municipal waste, sewage

http://dx.doi.org/10.1016/j.biortech.2014.09.143
0960-8524/Ó 2014 Elsevier Ltd. All rights reserved.
244 N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255

sludge, manure and black water, and showed high correlation with ization detector and a TR-FFAP capillary column with nitrogen car-
experimental results (El Fadel et al., 2012; Kerroun et al., 2010; rier gas.
Wichern et al., 2009; Lee et al., 2009; BouBaker and Ridha, 2007; COD (total and soluble) was determined using the modified
Jeong et al., 2005). Invariably, those applications rely on a set of 5220D procedure of Standard Methods for the Analysis of Water
data pertaining to one application using the final output (methane) and Wastewater (APHA et al., 1998) by HACH high-range COD kit
to calibrate the model. (HACH Company, Loveland, Colorado). Solid content (total, sus-
In this work, the ADM1 was applied for the first time to simu- pended, dissolved and volatile) was analyzed using Standard Meth-
late the thermophilic AD of food waste using the performance data ods (APHA et al., 1998) 2540B and 2540E procedures. Partial and
of four lab scale digesters. The simulation involved, in conjunction total alkalinities (PA and TA, respectively) were measured by titra-
with a sensitivity analysis, parameter optimization and cross- tion with HCl (0.2 N) to pH of 5.75 and 4.3, respectively. Total
validation against independent experimental data sets both at ammonia content was determined by spectrophotometry using
the intermediary by-products (pH, acetate, ammonia and Total HACH high-range ammonia kit (HACH Company, Loveland,
COD), and final experimental output (CH4-based results). To the Colorado).
best of our knowledge, no similar work has been reported in the Carbon and nitrogen content in the feed were determined using
literature. an organic elemental analyzer (Flash-1112 series EA, Thermo Finn-
igan). Total organic carbon in the feed was measured by combus-
tion catalytic oxidation/NDIR technique using a dedicated TOC
2. Methods
analyzer (TOC-VCSH Total Organic Carbon Analyzer, Shimadzu Cor-
poration, Japan). Total Phosphorous in the feed was measured
2.1. Experimental program
using inductively coupled plasma mass spectrophotometry (Agi-
lent 7500ce, EPA 200.8-3050).
Two anaerobic reactors (Bioflo 110, New Brunswick Scientific
Co.) of 14 L capacity (9 L working volume) were fed with food
waste collected from households and food markets, ground and 2.3. Simulation process
homogenized with a lab food grinder, and characterized. The raw
waste had a total solids content of 62–75 kg/m3 and a COD of The Anaerobic Digestion Model (ADM), developed by the Inter-
96–120 kg/m3. national Water Association (IWA), was adopted in this study. The
Both reactors were operated at 55 °C and were continuously model ensures mass balance amongst various bacterial communi-
mixed with an internal impeller at 80 rpm. The digesters were ties interacting with a heterogeneous substrate (Eqs. (1) and (2)) to
fed with an equal batch of waste 3 times per week (Monday, represent transformation processes within the reactor boundaries.
Wednesday and Friday) resulting in the lowest daily loading rate X
dSliq;i Q  
over the weekend (Friday, Saturday and Sunday). Accordingly, a ¼ : Sin;i  Sliq;i þ qj ti:j ð1Þ
dt V liq
harmonic behavior of daily methane generation was observed with j¼119

the lowest value being after Friday’s feeding (i.e., on Saturday, Sun-
where Sliq,i = reactor concentration of soluble state variable i;
day and Monday). Wasting occurred at a volume of 700 ml to
Q = flow into and out of the reactor, m3/day; Vliq = liquid reactor vol-
achieve a weekly average HRT of 30 days. The experiment lasted
ume, m3; Sin,i = input concentration of soluble state variable I;
for 700 days and two separate sets of experimental data, with sta-
Rqjmi,j = the sum of the specific kinetic rate (qj) of process j multi-
ble average weekly feeding rate, were selected for model calibra-
plied by the stoichiometric coefficients (vi,j).
tion and validation:
Set 1 – Digesters A and B were run at a stable organic loading dX liq;i Q   X
rate (OLR) of 2.4 gVS/L/d for 115 days (days 376–491) and ¼ : X in;i  X liq;i þ qj ti:j ð2Þ
dt V liq j¼119
referred to as A1 and B1, respectively.
Set 2 – The digesters were run at a constant OLR of 2 gVS/L/d for where Xliq,i = reactor concentration of particulate state variable i;
58 days (days 569–627) and the results are referred to as A2 and B2 Xin,i = input concentration of particulate state variable i.
for digesters A and B, respectively. In addition, acid-base equilibrium (Eq. (3)) is incorporated to
simulate pH temporal profile and associated potential inhibition:
2.2. Experimental analysis
Sac Spr Sbu Sva
SCatþ þ SNHþ þ SHþ  SHCO3      SOH  SAn ¼ 0
The temperature control was automated through a control unit
4 64 112 160 208
connected to a built-in temperature probe and a heating blanket. ð3Þ
The pH was continuously measured using a submerged pH probe
where Sac , Spro , Sbu , Sva , SNH4 and SHCO3 are the concentrations of
and controlled by manual addition of NaOH (5 M) solution – when
ionized forms of buffer components.
needed. Biogas measurements were taken once or twice daily,
Also, liquid–gas transfer processes are considered (Eq. (4)) to
using the water displacement method for total gas yield and a dual
determine biogas generation and composition:
wavelength infrared cell with reference channels (GEM-2000 mon-
itor, Keison Products, UK) for CH4 and CO2 concentrations (% by dSgas;i qgas V liq
volume). Biochemical analysis was conducted on a weekly basis ¼ Sgas;i þ q ð4Þ
dt V gas V gas T;i
whereby a fraction of the wasted material was used to measure
total and volatile solids and total COD. Then, samples were centri- where Sgas,i = gas phase concentration of gas component i; qgas = gas
fuged (4700 rpm, 10 min) and passed through filters with 1.2 lm flow outside the reactor; Vliq = reactor liquid volume; Vgas = reactor
openings for analysis of soluble solids, soluble COD, ammonia, gas volume (headspace volume); qT,i = specific mass transfer rate
and alkalinity. For VFAs, the filtrate was further passed through of gas i.
0.2 lm syringe filter and acidified with phosphoric acid to pH of The set of governing equations is solved simultaneously to sim-
2. The concentrations of acetate, propionate and butyrate inside ulate the dynamics and biological kinetics in the reactors. Details
the digesters were measured using a gas chromatograph (Trace about model governing equations, input parameters, and underly-
GC Ultra, Thermo Electron corporation) equipped with a flame ion- ing assumptions are described in Batstone et al. (2002).
N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255 245

The ADM1-based code using the MatlabÒ/SimulinkÒ platform calibration and validation of AD models (Souza et al., 2013; Zhou
(Rosen and Jeppson, 2006) was modified to allow for varying daily et al., 2012; Jeong et al., 2005).
flow rates. A separate module was incorporated to generate the
ADM1 input from experimental feed parameters using a transfor-
mation matrix (Zaher et al., 2009). Initial conditions in the reactors 3. Results and discussion
were assumed to be similar to those reported for the AD of food
waste (Bajzelj, 2009). The model’s biochemical, physical, and phys- 3.1. Experimental results
ico-chemical parameters were categorized through a sensitivity
analysis to select the most influential ones in subsequent optimiza- At first, the digesters were seeded with manure only and diges-
tion. For this purpose, the model was extended with a loop that ter A was run at 100 rpm mixing speed whereas digester B was run
calls for each parameter independently and assigns to it a range without mixing. The startup process lasted about 190 days during
of values, while keeping all other parameters constant. The which the Organic Loading Rate (OLR) was gradually increased to
assigned values were stored separately in an Excel sheet and ran- 2.4 gVS/L/d in digester A and 1.9 gVS/L/d in digester B, causing sys-
ged between 90% and 300% of benchmark values reported by tem overload in both digesters. Details of the startup procedure
Rosen and Jeppson (2006). Five performance indicators were and results are presented in Ghanimeh et al., 2012.
adopted in the analysis of the simulation results namely, daily After about 60 days of reduced feeding, both digesters were
methane generation rate (L/d), pH level, acetate (mg/L), ammonia reseeded. Digester A was inoculated with a control seed mix con-
(mg/L), and total COD (mg/L). In addition, a sensitivity analysis sisting of wasted digestate, cattle manure, and activated sludge.
for an equally weighted combination (EWC) of the five indicators Digester B was inoculated with the control mix and additional
was also conducted. The Sensitivity Index (SI) for the performance landfill leachate and municipal waste compost. The OLR was
indicators was calculated using Eq. (5a) (Souza et al., 2013) while slowly increased to 2.4 gVS/L/d, over 230 days, with continuous
for the EWC Eq. (5b) was used: mixing at 80 rpm. The digesters were run under steady loading
8P between day 376 and day 491. During this period, the specific
>
>
jSfixed ðtÞSnew ðtÞj
ðaÞ
>
< N methane generation was relatively stable at an average of 330
SI ¼ XNp P ð5Þ and 370 L CH4/kg VS in digesters A and B, respectively. The total
jSfixed ðtÞSnew ðtÞj
>
> 1
X ðbÞ concentration of VFAs was low (average = 510 and 220 mg Ac/L
> Np
: i¼1 Sfixed ðtÞ
in A and B, respectively) and average alkalinity was 4600 mg/L in
where Sfixed(t) and Snew(t) are the results of using a benchmark digester A and 5300 mg/L in digester B. The OLR was subsequently
(fixed) value of a parameter and the new results of varying the same reduced to about 2 gVS/L/d and maintained constant between day
parameter, respectively, calculated at time (t); N is the number of 569 and day 627. During this period, average specific methane gen-
days and Np is the number of performance indicators. eration was closely similar in both digesters with an average of
For the EWC, the SI was standardized because the five perfor- 350 L CH4/kg VS. The VFAs level was 770 mg/L and 230 mg/L, on
mance indicators have different units and as such Eq. (5b) was average, in digesters A and B, respectively. Afterwards, the digest-
ers underwent an organic shock for one week, followed by inter-
used. For example, the pH is dimensionless and ranges between
0 and 14 while the other indicators do not have an upper bound. rupted feeding and gradual incremental loading (Ghanimeh et al.,
2012).
So, for an effective sensitivity analysis and later on optimization,
all indicators should lie on the same scale.
The parameters to which the simulation results exhibited most
3.2. Sensitivity analysis
sensitivity were then optimized with the experimental data by
minimizing the error between simulated and experimental output
The sensitivity analysis of reactors A1 and A2 yielded similar
using the least square error method (Eqs. (6a) and (6b)). The error
patterns as B1 and B2, respectively, for all measured performance
was calculated automatically by applying the fminsearch function
indicators. Therefore, only the results of A1 and A2 are reported
in MatlabÒ on the results of two experiments (A1 and A2).
8 N in this paper. Results for B1 and B2 are provided in the accompany-
>
> X ing electronic appendix.
>
>
> ½Oi ðtÞ  Si ðtÞ2 ðaÞ While some parameters are highly sensitive at one end of the
< i¼1
Least Square Error ¼ ( ) ð6Þ simulated range, they were less sensitive at the other end. Taking
>
> XNp X N h i 2 the methane indicator for example, its results exhibited the great-
>
> Oi ðtÞSi ðtÞ
ðbÞ
>
: Oi ðtÞ est sensitivity to the inhibitory free ammonia concentration for
j¼1 i¼1
acetate degraders (KI NH3 ) and the rate of uptake of propionate
where Oi is the observed or measured value at time t, Si is the ADM1 (Km_pro) at the low KI NH3 and Km_pro values and a minor influence
simulated value corresponding to Oi at time t, and N is the number at the high KI NH3 and Km_pro values within the range used
of points in the data set. (Fig. 1a). In contrast, methane results are highly sensitive at the
Eq. (6a) was adopted for the five performance indicators while high range of the decay rates for propionate (Kdec_Xpro) and hydro-
for the EWC; Eq. (6b) was used for the same already mentioned gen (Kdec Xh2 ) degraders and practically insensitive at the lower end
reason. (Fig. 1b). If such parameters are considered for optimization, the
Finally, after the sensitive parameters were optimized for the model might be biased towards the extreme at which the parame-
various performance indicators (intermediary and final outputs) ter is most influential. As such, the other parameters are forced in a
on experiments A1 and A2, model validation was conducted using direction to counterbalance the extreme deviation from bench-
those same parameters on intermediary outputs of experiments B1 mark values leading to a pseudo-optimum solution thus limiting
and B2. the flexibility of the system by restricting some parameters to
The daily methane generation rate was selected to validate the either side of benchmark values. Accordingly, the parameters
model taking into consideration that the sensitivity of a parameter selected for further analysis and optimization are those showing
and its influence on the model depend on the number of mecha- equally high sensitivity at both extremes of the range (Fig. 1c),
nisms associated with it (Jeong et al., 2005). Knowing that methane which secures flexibility and provides real optimization in either
production is related to most processes, it is often used for direction of the benchmark value.
246 N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255

Table 2
Sensitivity Index (SI) 3 Summary of combined sensitive parameters of Set 1 and 2.

2 Methane pH Acetate Ammonia TCOD EWC

1 f h2 su fac_su fpro_su fpro_su fbu_su f h2 su


fpro_su Km_su fac_su fac_su fpro_su fpro _su
0 fac_su Km_ac Km_aa Km_su fac_su fac_su
0 100 200 300 400 Km_ac Ks_ac Km_ac Km_aa Km_ac Km_ac
Fraction of benchmark value (%) Ks_ac Kdec_xsu Ks_ac Ks_aa Ks_ac Ks_ac
(a) KI_NH3 Km_pro
Kdec_xsu Kdec_xaa Kdec_xsu Km_ac Kdec_xsu Kdec_xsu
Y_su Y_su Kdec_xaa Kdec_xsu Y_su Y_su
Y_aa Y_aa Y_su Kdec_xaa Y_aa Y_aa
Y_ac Y_ac Y_aa Y_su Y_ac Y_ac
Sensitivity Index (SI)

0.5 Y h2 Y h2 Y_ac Y_aa Y h2 Y h2


0.4 Y h2 Y_ac
0.3 Y h2
0.2
0.1
0 A high similarity was observed in simulating A1 and A2 exper-
0 100 200 300 400 imental results. Thus, a combined set of sensitive parameters was
Fraction of benchmark value (%) developed (Table 2) and used for subsequent analyses and optimi-
(b) Kdec_Xh2 Kdec_Xpro zation. Consistent with the findings of Jeong et al. (2005), simula-
tions based on methane exhibited high sensitivity to the yield of
catabolism of hydrogen (f h2 su ), propionate (fpro_su) and acetate
Sensitivity Index (SI)

4 (fac_su) from sugars, in addition to some anabolism yields (Ysu,


3 Yaa, Yac, Yh2 ), the Monod maximum specific uptake rate from ace-
2 tate (Km_ac), and the half saturation constant for acetate (Ks_ac).
1 Similarly, the simulations exhibited sensitivity to the decay rate
0 of sugar degraders (Kdec_Xsu), which was one of the prominent fac-
0 100 200 300 400 tors in the thermophilic digester in Lee et al. (2009) rather than the
Fraction of benchmark value (%) mesophilic one (similar to our case). Note that Batstone et al.
(c) Km ac Ks ac (2002) reported low sensitivity to the yields of anabolism and
kinetic parameters. The simulation results using acetate-based
Fig. 1. Sensitivity index of selected parameters (taking methane as indicator): (a) analysis were equally sensitive to the same parameters defined
Inhibitory free ammonia concentration for acetate degraders (KI NH3 ) and rate of for the methane-based best fit with the exception of the yield of
uptake of propionate (Km_pro); (b) decay rates for propionate (Kdec_Xpro) and
hydrogen catabolism (f h2 su ), the Monod maximum uptake rate
hydrogen degraders (Kdec h2 ); (c) uptake rate for acetate (Km_ac) and half saturation
constant for acetate (Ks_ac). for amino acids (Km_aa), and the decay rate of amino acids degrad-
ers (Kdec-Xaa). The latter two parameters did not cause high sensi-
tivity in systems treating a mixed waste of flour and dog food
Therefore, sensitivity indices at both extremities of the simu- (Lee et al., 2009). When examining the sensitivity results of the
lated range (90% and +300% away of benchmark values) were cal- EWC, it is readily noticeable that they are the same as that of the
culated separately and the parameters were ranked according to methane optimization, hence it is expected that optimization using
their respective Sensitivity Index (rank 1 for highest SI and rank this approach will fine-tune the methane parameters to capture
52 for lowest SI) for both extremes of the range. Then, the param- the trend of other performance indicators as well. In short, the
eters were prioritized based on the average of their ranks at both results indicate that sensitive parameters differ depending on asso-
extremities and the parameters with an average SI less than or ciated processes (Table 2) which infers that using methane gener-
equal to 10 were selected (Table 1). ation in the optimization process may not produce the best fit
solution for all other performance indicators.

Table 1
Parameters of high sensitivity with respect to all performance indicators and EWC.

Sensitive parameters Average SIa ranking of sensitive parameters


Methane pH Acetate Ammonia COD (Total) EWC
A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2
fac_su 6 4 8 6.5 9 3 7 8 8.5 5 5.5 3
fpro_su 8 7 – – – 9.5 10 8 7.5 7.5 – 7
fbu_su – – – – – – – – 10 8.5 – –
f h2 su – 5 – – – – – – – – – 9.5
Km_su – – – 7 – – 6.5 6 – – – –
Km_aa – – – – 7.5 7 6 5.5 – – – –
Ks_aa – – – – – – 10 8.5 – – – –
Km_ac 2.5 3 2.5 3 3 2 8.5 7 2 2 3 2.5
Ks_ac 7 5.5 8.5 9.5 8.5 9 – – 6.5 5.5 6.5 6
Kdec_xsu 6.5 – 3.5 5.5 6 6.5 – 5.5 6.5 10 6 –
Kdec_xaa – – – 7.5 8 8.5 7.5 7 – – – –
Y_su 2.5 2 1 1 3 3 3 2 2 1 2 1.5
Y_aa 9 – 7 – 9 8.5 8.5 9.5 10 – 8 –
Y_ac 6 9 7.5 – 4.5 6 7 – 5.5 – 5 –
Y h2 7 7 8.5 – 8.5 7 9 – 6.5 6 7 7.5
a
Average of SI at 90% and +300% away from benchmark values (SI = sensitivity index).
N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255 247

3.3. Model calibration: optimization based on methane where ADM1 performs best, were not fully achieved thus causing
the A2 parameters to deviate more from the benchmark values.
The results of the optimization process for A1 and A2, based on This suggests that the initial values reported by Rosen and
the methane generation best fit, are shown in Table 3 and com- Jeppson (2006) are ideal for conditions similar to that of A1 and
pared to the literature. The calibrated parameters of A1 are closely for methane-based optimization as will be demonstrated when
similar to the base parameters reported by Rosen and Jeppson conducting optimization based on other performance indicators.
(2006) with a deviation ranging between 0.7% and 30.0% in abso- A comparison of the average absolute percent relative error
lute value while a larger offset is observed for digester A2 with a between the five component indicators as a result of methane opti-
deviation ranging between 6.2% and 98.5% in absolute value. A mization is summarized in Table 4. Note that methane, acetate and
similar high deviation (259%) from the values given in Rosen and TCOD simulation results exhibited an error that is within similar
Jeppson (2006) was observed for Ks_ac in Jeong et al. (2005). The ranges reported in the literature (Jeong et al., 2005; Lee et al.,
results of both digesters showed that the yield of sugars Y_su is 2009; Wichern et al. 2008). However, the error in the simulated
the most sensitive parameter with the highest deviation of pH was higher than that reported by Wichern et al. (2008) which
98.5% (outside the sensitivity range of 90% to 300%) in A2. can be attributed to different experimental temperatures (thermo-
Fig. 2 displays the plots of the 5 performance indicators of A1 philic in this study vs mesophilic by Wichern et al. (2008)). It is
based on methane optimization. The model predicts methane gas worth noting that the high absolute percent relative error in
generation adequately and follows closely the experimental pat- ammonia for digester A1 is due to the huge difference between
tern throughout the experiment duration (Fig. 2a). Even though the model and experimental results at the beginning of the simu-
the model under predicts the acetate level, it captures the general lation, between day 0 and 40. If we discount these points, the
trend and depicts the presence of a sudden jump, around day 95 110% error on ammonia for A1 (Table 4) would be 29.67%. Also,
(Fig. 2c), which is of major importance for system stability and con- the high discrepancy in acetate results may be attributed to the
trol. Similarly, the pH was generally underestimated; yet the sim- low number of experimental data points. For instance, the high
ulated plot follows the quasi-stable experimental behavior and average error for A2 (54%, Table 4) is mainly due to an outlier on
clearly captures the alarming decrease, on day 95 (Fig. 2b), coincid- day 29. If only this point is omitted, the average error drops to 14%.
ing with the sudden increase in acetate. As to the ammonia level,
the simulated results started off the true values but converged, 3.4. Model calibration: optimization based on other performance
after day 50, to coincide with experimental data (Fig. 2d). The pat- indicators
tern of total COD, consisting of an increasing curve followed by a
stabilizing period, was captured but with a considerable offset In an attempt to further improve the calibration results, the
from real values (Fig. 2e). optimization process was repeated for A1 and A2 using major per-
Fig. 3 displays the plots of the 5 performance indicators of A2, formance indicators other than methane, including pH, ammonia,
based on methane optimization. Similar to A1, the daily methane acetate, TCOD and finally an EWC of the five performance indica-
generation was captured (Fig. 3a). Compared to A1, the pH results tors. The results showed a greater deviation from the initial values,
were closer to real values with a minimal offset beyond day 20 in comparison with methane-based optimization (Table 5). For
(Fig. 3b). Similarly, the simulated acetate concentrations coincide example, acetate-based optimization showed that the deviation
with real values, where provided, except for day 29 where the of the parameters ranged between 2.2–113.5% and 6.3–307.5%
model shows a steep peak (Fig. 3c). This can be either a false alarm- (absolute value) for A1 and A2, respectively, compared to
ing output or a true acetate surge that was not captured by exper- 0.4–30.0% and 12.2–98.5% for A1 and A2, respectively, in meth-
imental data, due to the low sampling frequency, and occurring ane-based optimization, with TCOD and ammonia exhibiting the
during a period (between days 25 and 35) of relatively low pH widest deviation ranges. As for the EWC, the deviation was some-
(Fig. 3b). Similar to A1, the error in ammonia predictions was where between that of methane and other parameters and ranged
decreasing with time (Fig. 3d) but did not reach convergence, most between 0.5–77% for A1 and 11.3–132% for A2, reflecting the effect
probably because of the short experimental duration (50 days of combining all performance indicators.
compared to 114 days for A1). Unlike A1, the TCOD results of A2 The simulated results (Figs. 4 and 5) showed that optimization
showed good correlation with experimental data (Fig. 3e). based on various performance indicators (with the exception of
The simulated methane pattern for A1 is harmonic with a rela- ammonia) is able to capture the general experimental trend, but
tively constant period and amplitude (Fig. 2a) which fairly agrees with a higher error compared to methane-based optimization
with the feeding pattern mentioned in the materials and methods curves (Table 6a). The second-best results were those of acetate-
section. While for A2, the simulated pattern (Fig. 3a) is inconsistent based optimization. In contrast, ammonia-based optimization
with its feeding pattern indicating that steady-state conditions, yielded unrealistic results and wrong overall patterns.

Table 3
Estimated parameters for digesters A1 and A2.

Parameter Initial values Rosen and Jeppson (A1) (% deviation from initial (A2) (% deviation from initial Jeong et al. (2005) Lee et al. (2009)
(2006) value) value)

f h2 su 0.19 0.2 (5.26%) 0.178 (6.23%) – –


fpro_su 0.27 0.269 (0.37%) 0.033 (87.8%) 0.54 –
fac_su 0.41 0.381 (7.1%) 0.374 (8.86%) 0.202 –
Km_ac 8 9.313 (16.4%) 14.41 (80%) – 14
Ks_ac 0.15 0.151 (0.67%) 0.1215 (19%) 259 0.6
Kdec_xsu 0.02 0.0226 (13%) 0.0245 (22.6%) – 0.03
Ysu 0.1 0.07 (30%) 0.001 (98.5%) 0.05 –
Yaa 0.08 0.085 (6.25%) 0.1 (25.6%) – –
Yac 0.05 0.0497 (0.6%) 0.0434 (13.6%) 0.1 –
Y h2 0.06 0.058 (3.33%) 0.1 (68.1%) 0.0282 –

The percent deviation is shown in bold.


248 N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255

Fig. 2. Simulated and experimental data for A1 digester based on methane optimization: (a) methane; (b) pH; (c) acetate; (d) ammonia; (e) TCOD.

On the other hand, optimizing on each performance indicator improved average results compared to optimization based only on
improves its respective errors, as shown in Table 6 (a). For exam- methane. In fact, this approach exhibited major improvements for
ple, pH-based optimization reduced the error in pH simulation A2, namely in ammonia (16.5%), acetate (11%) and TCOD
results from 9.4% to 1.2% for A1 and from 5.7% to 2.2% for A2. These (10%), with a minor increase in error in methane (from 14% to
findings are further verified in Figs. 4 and 5, where optimization on 14.6%) and pH (from 5.7% to 6.6%). Also, all performance indicators
each performance indicator brings the simulated results closer to for A1 were moderately improved, with the exception of ammonia
the real pattern and values of that specific indicator, for both (5% increase in error), with the highest improvements being in ace-
digesters A1 and A2. Hence, optimizing for methane, even though tate (7%) and TCOD (3%). The higher overall percentage of
it yields realistic patterns, does not guarantee the optimum solu- improvement in A2 over A1 can be attributed to the higher devia-
tion for all outputs and thus it is critical to define the desired tion of the parameters from the benchmark. Yet, the decrease in
importance of each component prior to the optimization process. the error, in both experiments, could not be reduced to levels close
Alternatively, the EWC can provide a unified solution with overall to those determined separately for each performance indicator. In
N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255 249

Fig. 3. Simulated and experimental results of A2 digester based on methane optimization: (a) methane; (b) pH; (c) acetate; (d) ammonia; (e) TCOD.

Table 4
Average absolute percent relative error based on methane optimization (Standard deviation).

Methane pH Acetate Ammonia TCOD


A1 13.7% (10.94%) 9.4% (2.65%) 58.18% (27.45%) 110.2% (117.66%) 70.16% (43.42%)
A2 14% (15.67%) 5.74% (2.56%) 54.1% (98.21%) 310.63% (193.77%) 22.2% (15.8%)
Jeong et al. (2005)a 5–40% – 10–97% – –
Lee et al. (2009)a 5–35% – 5–200% – 5–90%
Wichern et al. (2008)b 5–25% 1–5% 5–200% – –

The percent deviation is shown in bold.


a
The relative percent error was not reported and thus approximated as a min–max range.
b
The error was reported but, for consistency, reported as a min–max range.
250
Table 5
Optimized parameters based on five performance indicators and EWC for A1 and A2.

Optimized value based on least square error in different performance indicators for A1 and A2 (% deviation from initial value)
Parameter Initial Methane pH Ammonia Acetate TCOD EWC
value
A1 A2 A1 A2 A1 A2 A1 A2 A1 A2 A1 A2
fac_su 0.41 0.381 0.46 (12.2%) 1.0211 (149%) 1.068 0.00134 1.3 (217.1%) 0.326 0.384 (6.3%) 0.0473 0.023 0.496 (21%) 0.6513
(7.1%) (160.5%) (99.6%) (20.5%) (88.46%) (94.4%) (58.85%)

N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255


fbu_su 0.13 - - - - - - - - 0.02 0.0312 - -
(84.62%) (76%)
fpro_su 0.27 0.269 0.037 - - 0.087 0.1406 0.2307 0.024 0.0118 0.0561 0.17 0.04273
(0.37%) (86.3%) (67.77%) (47.93%) (14.56%) (91.11%) (95.63%) (79.2%) (37.03%) (84.17%)
f h2 su 0.19 0.2 (5.26%) 0.1 (47.37%) - - - - - - - - 0.189 0.00733
(0.53%) (96.14%)
Km_su 30 - - 76.157 31.34 (4.46%) 4.36 34.16 - - - - - -
(153.85%) (85.5%) (13.87%)
Km_ac 8 9.313 4.23 (47.13%) 50.266 33.7 (321.3%) 78.03 5.24 7.824 4.89 153.43 18.9 (136.3%) 10.224 3.97650.3%)
(16.4%) (528.33%) (875.4%) (34.5%) (2.2%) (38.88%) (1817.9%) (27.8%)
Km_aa 50 - - - - 12.44 4.47 13.52 46.52 - - - -
(75.12%) (91.1%) (72.96%) (6.96%)
Ks_aa 0.3 - - - - 0.0612 0.238 - - - - - -
(79.6%) (20.67%)
Ks_ac 0.15 0.151 0.23 (53.3%) 0.0401 0.1405 - - 0.204 (36%) 0.338 0.076 0.8 (433.3%) 0.187 0.167 (11.33%)
(0.67%) (73.3%) (6.33%) (125.33%) (49.3%) (24.67%)
Kdec_xsu 0.02 0.0226 0.036 (80%) 0.011 (45%) 0.0219 (9.5%) 0.0192 (4%) 0.0421 0.0287 0.0251 0.008 (60%) 0.0024 0.0174 0.0464 (132%)
(13%) (110.5%) (43.5%) (25.5%) (88%) (13%)
Kdec_xaa 0.02 - - 0.00133 0.0565 0.0207 (3.5%) 0.0247 0.0427 0.0815 - - - -
(93.35%) (182.5%) (23.5%) (113.5%) (307.5%)
Y_su 0.1 0.07 0.00148 0.00144 0.0011 0.17 (70%) 0.002 (98%) 0.0655 0.00108 0.00366 0.002 (98%) 0.0534 0.001824
(30%) (98.52%) (98.56%) (98.9%) (34.5%) (98.92%) (96.34%) (46.6%) (98.18%)
Y_aa 0.08 0.085 0.0314 0.00321 0.0426 0.0733 0.22 (175%) 0.0561 0.05 (37.5%) 0.001 0.00175 0.1415 0.0104 (87%)
(6.25%) (60.75%) (96%) (46.75%) (8.375%) (30%) (98.75%) (97.8%) (76.9%)
Y_ac 0.05 0.0497 0.0603 0.00584 0.0133 0.192 (284%) 0.167 (234%) 0.0359 0.06 (20%) 0.0042 0.0103 0.0281 0.104 (108%)
(0.6%) (20.6%) (88.32%) (73.4%) (28.2%) (91.6%) (79.4%) (43.8%)
Y h2 0.06 0.058 0.026 0.0094 0.193 0.0035 0.46 (666.7%) 0.0662 0.0288 (52%) 0.0075 0.0028 0.0667 0.024 (60%)
(3.33%) (56.67%) (84.3%) (221.7%) (94.17%) (10.33%) (87.5%) (95.3%) (11.16%)

The percent deviation is shown in bold.


N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255 251

Fig. 4. Model optimization results of A1 based on 5 performance indicators and EWC: (a) methane; (b) pH; (c) acetate; (d) ammonia; (e) TCOD.

A1, the equally weighted combination reduced the error on pH, - The overall pattern of daily methane generation can be cap-
acetate and TCOD from 9.4%, 58.2% and 70.2% (from methane based tured with optimization on any of the indicators used in this
optimization) to 8.4%, 51.4% and 67.4% which are still higher than study, at the exception of ammonia. But accurate values are
the minimum values reached in calibration on separate indicators obtained only with methane- or acetate-based optimization.
(1.21%, 40.8%, 49.6%). Similarly for A2, EWC calibration dropped - The simulated results of ammonia start at a high level and con-
the error on ammonia, acetate and TCOD from 310.6%, 54.1% and verge to real values with time.
38.8% (from methane based optimization) to 8.4%, 51.4% and - Ammonia-based optimization is not able to capture the pattern
67.4% which are still higher than the minimum values reached in of any indicator, other than ammonia, and results in high errors.
calibration on separate indicators. However, the equally weighted - Multi-objective minimization improves the methane-based cal-
combination remains a reliable approach since it reduces the errors ibration results and provides a unified set of parameters equally
on methane-based optimization and provides a unified set of pre- accounting for each performance indicator.
dictors that equally account for all performance indicators.
Finally, the following general observations can be drawn: 3.5. Model optimization: limitations

- Methane- and acetate-based optimization yield very close Despite the improvements made on ammonia and TCOD, both
results for all indicators. graphs (Fig. 5d and e, respectively) do not capture the trend and
252 N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255

Fig. 5. Model optimization results of A2 based on five performance indicators and EWC: (a) methane; (b) pH; (c) acetate; (d) ammonia; (e) TCOD.

tend to have the same shape irrespective of the indicator used for actual acetate. Thus selecting the starting point in the acetate opti-
optimization. One possible reason could be that a micro fraction of mization process to match the parameters from the methane-
the entire sample used for measuring acetate, ammonia and TCOD based optimization will bypass the setback of having a low number
concentrations is not representative of the entire digester. of data points and simplify optimization to fine tuning. For digester
In comparison, one potential reason that the pH optimization A2, this has worked perfectly by reducing the error from 54% to
resulted in a significant improvement of pH estimates, despite 3.2%. However, for A1, the low number of data points and high pos-
the fact that measurements were also taken at a single location, sibility of the data points not being representative of the digester
was the presence of numerous experimental points which trans- and improved methane optimization from 58% to 40% only. In fact,
formed the problem to a data-fitting one. In other words, the high the non-steady, wave like patterns of measured points could have
repetition rate of pH measurements at single point, coupled with made it difficult for the optimizer to capture the trend starting
mixing provided the optimization problem with a more represen- with Batstone’s suggested values.
tative dataset to optimize with. For the cases of ammonia and TCOD, setting the starting point
As for the case of acetate optimization, despite the low number equal to methane-based optimization or any other indicator-based
of data points (less than the number of parameters), the trend was optimization did not significantly improve the result since none of
fine-tuned from that of methane optimization due to the fact that the indicator-based optimizations (other than ammonia) was able
methane is a direct derivative of acetate. In other words, the trend to capture the actual trend. Notice for the case of ammonia, that all
generated from methane optimization more or less replicates the component-based optimization have huge offsets between day 0
Table 6
Average absolute percent relative error based on 5 performance indicators for: (a) optimization on digesters A1 and A2; (b) validation on digesters B1 (using A1 results) and B2 (using A2 results) and (c) validation on digesters B1 (using
A2 results) and B2 (using A1 results).

Absolute Average offset from experimental value (%) (Standard deviation)


Performance Methane pH Ammonia Acetate TCOD
indicator*
(a) Error for optimization on digesters A1 and A2
A1 A2 A1 A2 A1 A2 A1 A2 A1 A2
Methane 13.7% (10.94%) 14% (15.67%) 9.4% (2.65%) 5.74% (2.56%) 110.2% (117.66%) 310.63% 58.18% (27.7%) 54.1% (98.21%) 70.16% (43.4%) 38.83% (15.8%)
(193.8%)
pH 64.62% 113.3% (46%) 1.21% (1%) 2.23% (1.9%) 149.76% 292.75% (198%) 87.84% (9.16%) 85.1% (4.06%) 53.27% (38%) 23.58% (14.1%)
(30.71%) (113.34%)
Ammonia 89% (19.6%) 95.8% (13.3%) 34.6% (13.37%) 57.47% (18.5%) 100% (108%) 218.8% (195%) 1774.9% (1022.5%) 1059.9% (740.5%) 150% (100.8%) 134.28%
(63.84%)
Acetate 15.55% (9.53%) 14.8% (19.5%) 8.5% (2.32%) 5.54%2.53%) 118.17% (118.7%) 314.5% (196.1%) 40.85% (39.3%) 3.22% (3.92%) 66.38% (42.3%) 19.1% (14.32%)
TCOD 55.61% (7.92%) 40.6% (13.3%) 6.4% (1.48%) 5.76% (2.81%) 159.11% (112.8%) 329.75% (192%) 97.15% (1.5%) 100.52% (133.5%) 49.6% (37.1%) 12% (10.71%)
EWC 13.6% (11.5%) 14.6% (16.56%) 8.4% (2.4%) 6.62% (2.55%) 115.4% (118.11%) 294.31 (194.7%) 51.4% (26.33%) 43.02% (12.21%) 67.4% (42.55%) 29.34% (19.41%)

N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255


(b) Error for validation on digesters B1 (using A1) and B2 (using A2)
B1 B2 B1 B2 B1 B2 B1 B2 B1 B2
Methane 15.22% (15.4%) 21.8% (10.7%) 13.33% (2.1%) 12.5% (1.27%) 181.47% 434.3% (384.4%) 62.06% (16.93%) 67.23% (58.7%) 112.88% (47.3%) 47.5% (31.38%)
(196.16%)
pH 65.61% 72.9% (31%) 4.38% (1.3%) 7.26% (1.3%) 237.77% (196%) 410.84% 88.46% (5.5%) 78.36% (14%) 91.8% (41.66%) 48.1% (26.37%)
(33.08%) (379.4%)
Ammonia 88.86% (19.7%) 94.5% (18.6%) 37.71% 63% (15.76%) 172.18% 313.23% 2405.26% 1072.83% (922.1%) 205.33% (101.3%) 187% (95.76%)
(12.02%) (186.63%) (342.8%) (1901.3%)
Acetate 16.37% 27.5% (23.75%) 12.5% (1.84%) 13.86% (5.3%) 194.36% (197.3%) 440.5% (388.2%) 36.73% (17.37%) 448.6% (832.6%) 108.08% (46%) 47.48% (34.1%)
(14.22%)
TCOD 55.83% (9.4%) 52.14% (9.7%) 10.53% (1.26%) 12.42% (1.33%) 249.9% (197.07%) 459.6% (393.5%) 96.88% (2.3%) 202.67% (337.4%) 87.3% (40.58%) 32.32% (19.52%)
EWC 15.24% (15.8%) 44.82% 12.4% (1.9%) 20.69% (11.6%) 189.45% 420% (369%) 38.4% (24.17%) 950.3% (1415.5%) 109.49% (46.37%) 75.85% (68.82%)
(68.68%) (197.08%)
(c) Error for validation on digesters B1(using A2) and B2 (using A1)
B1 B2 B1 B2 B1 B2 B1 B2 B1 B2
Methane 44.42% 70.56% 9.97% (1.25%) 34.62% 229.91% 400.4% (356.9%) 35% (22.05%) 2747.4% (2958.7%) 147.52% (79.58%) 109.24% (91.2%)
(19.77%) (38.76%) (15.85%) (195.67%)
pH 37.57% (24%) 83.16% (37.4%) 5.84% (1.6%) 5.44% (1.47%) 209.82% 445% (388%) 79.48% (7.5%) 93.53% (2.58%) 136.07% (67.92%) 34.32% (18.74%)
(196.44%)
Ammonia 97.7% (9.82%) 87.22% (21.1%) 64.5% (16.87%) 38.1% (12.34%) 143.15% (171.2%) 359.83% 1802.4% (835%) 856.48% (639.22%) 317.25% 115.8% (85.5%)
(344.5%) (127.77%)
Acetate 42.04% (20%) 70% (39.17%) 9.96% (1.28%) 37.86% 232% (197.7%) 386.3% (389.8%) 28.52% (17.6%) 1714.1% (2032.4%) 139.5% (75.69%) 106.4% (93.57%)
(19.87%)
TCOD 54.3% (8.96%) 52.74% (9.58%) 11.08% (1.47%) 11.71% (1.41%) 249.28% 460.5% (394%) 205.3% (175.55%) 98.7% (0.83%) 89.1% (41.08%) 30.32% (92.26%)
(197.07%)
EWC 99.9% (0.34%) 70.61% 40.77% (1.48%) 34.07% 249.3% (197%) 410.28% 6619.71% 3261.47% 320.3% (120.66%) 107.88%
(38.84%) (15.95%) (361.3%) (2483.5%) (3544.5%) (92.26%)

The percent deviation is shown in bold.


*
The components highlighted in bold respectively indicate the basis upon which optimization and validation were conducted.

253
254 N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255

and day 40 from the experimental data due to the decaying curve. Despite the general increase in error from Table 6a (calibration
The situation is less severe for TCOD where all models seem to results) to Table 6b (validation results), the optimum approach for
some extent overestimate the experimental data throughout the modeling each performance indicator remains the same. This pro-
duration but capture the experimental trend, especially for diges- vides a strong validation of the optimization results with the only
ter A2. Hence, it is recommended to consistently take multiple exception being the acetate-based validation on digester B2, where
measurements of each indicator both spatially and temporally. it did not yield the lowest error. The reason could be attributed to
the possibility of over fitting when optimizing for the acetate
3.6. Model validation parameter on digester A2 making the parameters only valid for
that specific digester at those specific points, and not optimum
To validate the models of digester A in Sets 1 and 2, the param- for any other case.
eters optimized on the various performance indicators generated In addition to validating the results of A1 and A2 on B1 and B2
for A1 and A2 were used as an input in digesters B1 and B2. A sam- respectively (Table 6b), the optimized parameters from A1 and A2
ple of curves for validation of B1 using performance indicators of were applied to B2 and B1 respectively for further validation
A1 is shown in Fig. 6 and a summary of the errors of each perfor- (Table 6c). By comparing the ranking of the optimum approaches
mance indicator from applying the parameters of A1 on B1 and between Table 6b and Table 6c, it is evident that the two are not
A2 on B2 is shown in Table 6b. similar. This can be attributed to the fact that experiment set #1

Fig. 6. Model validation results of B1 using A1’s performance indicators and EWC parameters: (a) methane; (b) pH; (c) acetate; (d) ammonia; (e) TCOD.
N.M. Atallah et al. / Bioresource Technology 174 (2014) 243–255 255

(A1 and B1) and experiment set #2 (A2 and B2) were started with Standard Methods for the Examination of Water and Wastewater, 20th ed.,
Washington, DC.
different inocula, resulting in different microbial compositions and
Banks, C.J., Chesshire, M., Stringfellow, A., 2008. A pilot-scale comparison of
leading to major differences in key reaction rates and factors. mesophilic and thermophilic digestion of source segregated domestic food
Because the EWC approach equally involves all of the indicators, waste. Water Sci. Technol. 58, 1475–1481.
it is the most sensitive approach and thus displayed the highest Batstone, D.J., Keller, J., Angelidaki, I., Kalyuzhnyi, S.V., Pavaostathis, S.G., Rozzi, A.,
Sanders, W.T.M., Seigrist, H., Vavilin, V.A., 2002. The IWA Anaerobic Digestion
change. At the same time, EWC of A2 displayed higher change than Model No. 1 (ADM1). IWA publishing, London, UK.
EWC of A1 since it had higher improvement on the methane-based BouBaker, F., Ridha, B.C., 2007. Modelling of the mesophilic anaerobic co-digestion
approach and displayed a greater deviation in the parameters. of olive mill wastewater with olive mill solid waste using Anaerobic Digestion
Model No.1 (ADM1). Bioresour. Technol. 99, 6565–6577.
Cesur, D., 2004. Extension of Anaerobic Digestion Model No. 1 and Its
4. Conclusion Implementation, (Ph.D. dissertation). Colorado State University.
El Fadel, M., Maroun, R., Bou Fakher Eldeen, R., Ghanimeh, S., 2012. ADM1
performance using SS-OFMSW with non-acclimated. Water Sci. Technol. 66,
ADM 1 was used to simulate two separate experimental setups 1885–1892.
each involving two digesters (A and B) treating SS-OFMSW. The Esposito, G., Frunzo, L., Panico, A., Pirozzi, F., 2011. Modelling the effect of the OLR
and OFMSW particle size on the performances of an anaerobic co-digestion
sensitivity analysis on methane, pH, acetate, ammonia, TCOD and
reactor. Process Biochem. 46, 557–561.
equally weighted combination revealed that each component had Ghanimeh, S., Saikaly, P., El-Fadel, M., 2012. Impact of mixing on startup of
different influencing parameters depending on the processes anaerobic digesters treating OFMSW. Bioresour. Technol. 117, 63–71.
linked to it. It was found that methane optimization, a commonly Ghanimeh, S., El-Fadel, M., Saikaly, P., 2013. Improving the stability of thermophilic
anaerobic digesters treating SS-OFMSW through enrichment with compost and
adopted approach, does not adequately predict other intermediary leachate seeds. Bioresour. Technol. 131, 53–59.
outputs. However, the equally weighted multi-objective optimiza- Jeong, H.S., Suh, C.W., Lim, J.L., Lee, S.H., Shin, H.S., 2005. Analysis and
tion improved the results from that of methane-based optimiza- application of ADM1 for anaerobic methane production. Bioprocess Biosyst.
Eng. 27, 81–89.
tion. Validation certified the results obtained from the Kerroun, D., Mossaab, B.-L., Abdessalam Hassen, M., 2010. Use of ADM1 model to
optimization process with the exception of acetate based optimiza- simulate the anaerobic digestion process used for sludge waste treatment in
tion on A2 due to over-fitting. thermophilic conditions. Turk. J. Eng. Environ. Sci. 4, 121–129.
Lee, M.-Y., Suh, C.-W., Ahn, Y.-T., Shin, H.-S., 2009. Variation of ADM1 by using
temperature-phased anaerobic digestion (TPAD) operation. Bioresour. Technol.
Acknowledgements 100, 2816–2822.
Rosen, C., Jeppsson, U., 2006. Aspects on ADM1 Implementation within the BSM2
framework, Technical Report. Department of Industrial Electrical Engineering
This work was supported by the National Council for Scientific
and Automation, Lund University, Lund, Sweden.
Research, Lebanon. Special thanks are extended to the United Souza, T.S., Carvajal, A., Donoso-Bravo, A., Pena, M., 2013. ADM1 calibration
States Agency for International Development for its support in using BMP tests for modeling the effect of autohydrolysis pretreatment on
the performance of continuous sludge digesters. Water Res. 47, 3244–
acquiring the equipment used in the experimental program.
3254.
Ward, A.J., Hobbs, P.J., Holliman, P.J., Jones, D.L., 2008. Optimisation of the
Appendix A. Supplementary data anaerobic digestion of agricultural resources. Bioresour. Technol. 99, 7928–
7940.
Wichern, M., Gehring, T., Fisher, K., Andrade, D., Lubken, M., Koch, K., Gronauer, A.,
Supplementary data associated with this article can be found, in Horn, H., 2009. Monofermentation of grass silage under mesophilic conditions:
the online version, at http://dx.doi.org/10.1016/j.biortech.2014. measurements and mathematical modeling with ADM1. Bioresour. Technol.
100, 1675–1681.
09.143. Zaher, U., Buffiere, P., Steyer, J.-P., Chen, S., 2009. A procedure to estimate proximate
analysis of mixed organic wastes. Water Environ. Res. 81, 407–415.
References Zhou, H., Li, H., Wang, F., 2012. Anaerobic digestion of different organic wastes
for biogas production and its operational control performed by the
modified ADM1. J. Environ. Sci. Health A Toxic/Hazard. Subst. Environ.
APHA (American Public Health Association), AWWA (American Water Works and
Eng. 47, 84–92.
Protection Association) and WPCF (Water Pollution Control Federation), 1998.