Beruflich Dokumente
Kultur Dokumente
Abstract: This paper deals with the optimisation and cal- dynamics of the process components, and thus is adopted as
ibration of a two-reaction model of anaerobic digestion. A the primary operational strategy. To overcome the inherent
steady-state optimisation of the digester dilution rate was problem of instability, it is necessary to model, simulate and
carried out to obtain optimum values for bacteria concen- optimise such anaerobic digesters.
trations, which were used in the dynamic optimisation of The paper is organised as follows. The first section in-
the digester start-up process. Additional control of the volves modelling, simulation and dynamic optimisation of
digester alkalinity and a constraint on the methane frac- the process with the aid of GAMS and gPROMS. The sec-
tion were then added to refine the optimisation. The re- ond section focuses on enhancing the calibration of the
sults of 4 days of experiments in a 1-m3 digester were
model proposed by [1].
then used to improve the calibration of the model through
the conduct of a general sensitivity analysis to identify the
model parameters to be refined. MODELLING, SIMULATION AND OPTIMISATION OF
Keywords: dynamic optimisation; sensitivity analysis; AN ANAEROBIC DIGESTER
parameter estimation
Equations of a Dynamic Mass-Balance Model
S1 (g/L) or S2 (mmol/L)
5
tf t
SI f 10tf
0.6 4 p p 0.1
8
reach its desired steady-state value at the end of the time
horizon. The fluctuations observed in the second arc en- 6
sured that X1 and X2 increased gradually towards their de- 4
sired values at the end. In the last arc, the dilution rate was
2
zero since S1 and S2 had reached sufficient values to in-
crease X1 and X2 to their steady-state values. 0
1max 2max X1 X2
p
Sensitivity Analysis
Figure 3. Sensitivity index of the optimal solution to uncer-
As a large degree of uncertainty is typically associated with tainties (+10%) in the parameters/initial conditions.
bioprocess models, it is important to check how sensitive
ence between the two in terms of control values and control min tf
D ( t ), 0 t tf
structure, except the minimal time required, as evidenced by Z (t )
the close overlap between both profiles for most part of the
0.01
time horizon. This means that only slight changes need to be 0.01
made towards the end of the time horizon to further opti-
mise the control strategy in response to the uncertainty as- 40
s.t. Kr D Q F, (0)
sociated with 1max . This should not incur too significant a 0
0
cost in practice.
50
1.0 X 1 (tf ) 0.196
X 2 (tf ) 0.0702
0.8
0 D (t ) 1
0 Z (t ) 100
D (d1)
0.6
PC (t )
0.5 for all t , except t 0
PT (t )
0.4
The optimal solutions of the two controls, namely the di-
0.2 lution rate and alkalinity feed concentration, are shown in
Fig. 5, while the responses of the key variables, biomass
0.0 concentrations and methane fraction, are shown in Fig. 6.
0 10 20 30
t (days)
1.0 70
1max = 1.2 1max = 1.32
65
0.8
Figure 4. Optimal control strategy for 1max at nominal value
and 10% higher. 60
Zin (mmol/L)
0.6
D (d1)
55
Methane Composition of Biogas during Optimal
0.4
Start-Up 50
yM
0.08 0.4
Simulation of Experiment and Comparison with
Available Measurements
0.04 0.2
The model was simulated based on the influent conditions
0.00 0.0 reported in Table I and using the parameter values estimated
0 10 20 30 in [1]. The initial measurement values of S1, S2, C and Z
t (days) were used as initial conditions for the simulation, providing
X1 X2 yM four of the six initial conditions required by the model. The
Figure 6. System response to dynamic optimisation of dilu- remaining two initial conditions have been estimated by
tion rate and digester alkalinity. calculating the values of X1 and X2 that would produce re-
sponse trajectories most similar to those of the experimental
A comparison of these results with those obtained in Fig.
measurements. Fig. 7 represents the simulation results
2 indicates a slight difference in the behaviour of dilution
alongside the experimental data.
rate, due to the introduction of one new control and the me-
thane fraction constraint. The control arc had a similar over- 12 150
all shape, but the dilution rate took on the extreme values of 8 100
S2 (mmol/L)
S1 (g/L)
140
100
entirely different behaviour from that in Fig. 2, with a sharp 120
60 100
initial increase followed by a steady drop to reach the min-
20 80
imum value dictated by the imposed constraint. Apart from 0 1 2 3 4 0 1 2 3 4
t (days) t (days)
a sharp deviation between days 10 and 12, the methane frac-
7.6
tion was then maintained at this minimum value of 0.5 until 7.4
it increased slightly to 0.6 at the end of the time period. This
pH
7.2
7.0
is in contrast with the behaviour seen in Fig. 2, where the
6.8
methane fraction decreased to a minimum value of around 6.6
0 1 2 3 4
0.2 before increasing gradually to reach a maximum value t (days)
SIS2
0.0 2
1
-0.5
0
-1
-1.0
-2
-1.5 -3
0.3 0.5
0.0
0.0
-0.3
SIX1
SIX2
-0.5
-0.5
-0.8
-1.0 -1.0
1.0 0.5
0.8
0.6 0.0
0.4
-0.5
0.2
SIC
SIZ
0.0
-1.0
-0.2
-0.4 -1.5
-0.6
-0.8 -2.0
Figure 8. Sensitivity for the model parameters. The mean changes of S1, S2, X1, X2, C and Z are represented with respect to a
10% increase in the values of the kinetic, hydrodynamic and stoichiometric parameters.
Local Sensitivity Analysis of Response Trajectories parameter must be integrated over the time horizon in order
to obtain the sensitivity indices as follows:
In order to obtain better agreement between the model and t f y p
experimental data, a refinement of the experimental parame- 0 p y dt
ters can be carried out. However, it is important to first con- SI y tf
S2 (mmol/L)
8 100
k1 and k3. Five parameters have been selected as there are
S1 (g/L)
five quantities of which experimental measurements are 4 50
180 180
140 160
Z (mmol/L)
Parameter estimation and statistical analysis
C (mmol/L)
140
100
120
60
Using gPROMS, the five parameters as mentioned earlier 100
20 80
were re-estimated for the given set of experimental data. 0 1 2 3 4 0 1 2 3 4
t (days) t (days)
The results are presented in Table II.
7.4
Table II. Nominal values and new estimates of parameters.
7.1
pH
Parameter Unit Nominal value New estimate 6.8
20
Parameter Value 95% Confidence 95% t-value
Residuals
interval 0
0.167 0.0208 8.02 0 1 2 3 4
1max 0.360 0.0541 6.66 -20