Identiication of Algae Growth Kinetics

Systems and Control Group
Thesis Systems and Control
IDENTIFICATION OF
ALGAE GROWTH KINETICS
CAlg
Iin
Mochamad Bagus Hermanto
Iout
light path
May 2009
Light intensity [/mmol / ms]
1500
1000
0.1
500
0.09 0 2 4 6 8 10 12
Time [h]
0.08
1
0.07
Specific growth rate [/h]
0.5
0.06
Dilution [/h]
0.05 0
0.04 -0.5
0.03 -1
0.06 0 2 4 6 8 10 12
0.02 Time [h]
0.01
0.05
0
0 20 40 60 80 100 120 140 160 180 200
0.04 Light intensity [/mmol/ms]
0.03
0.02
0.01
0
0 20 40 60 80 100 120 140 160 180 200
Light intensity [/mmol/ms]
IDENTIFICATION OF
ALGAE GROWTH KINETICS
Name course : Thesis project Systems and Control

Number : SCO+80436
Study load : 36 ects
Date : May 2009
Student : Mochamad Bagus Hermanto

Registration number : 82+08+05+329+110
Study programme : MAB (Agricultural and Bioresource
Engineering)
Supervisors : dr.ir. A.J.B. Boxtel

dr.ir. K.J. Keesman
Examiners : Prof.dr.ir. G. van Straten,
Group : Systems and Control Group
Address : Bornsesteeg 59
6708 PD Wageningen
the Netherlands
Tel : +31 (317) 48 21 24
Fax: +31 (317) 48 49 57
Title page
Identification for Algae Growth Kinetics
Table of content
Title page........................................................................................................................................ i
Table of content..............................................................................................................................ii
Index of figures ..............................................................................................................................iii
Index of tables ...............................................................................................................................iv
Acknowledgement ......................................................................................................................... v
Summary .......................................................................................................................................vi
1. Introduction ........................................................................................................................... 1
1.1. Problem definition ........................................................................................................ 2
1.2. Research objectives..................................................................................................... 3
1.3. Research approach...................................................................................................... 3
1.4. Thesis outline ............................................................................................................... 3
2. Algae photobioreactors......................................................................................................... 4
2.1. Photobioreactor model................................................................................................. 4
2.2. Kinetic expression ........................................................................................................ 6
2.3. Light attenuation........................................................................................................... 7
3. Optimal input design ............................................................................................................. 8
3.1. Parametric sensitivity ................................................................................................... 8
3.2. Cost function .............................................................................................................. 10
3.3. Dynamic optimisation ................................................................................................. 10
3.4. Analytical solution ...................................................................................................... 12
3.5. Numerical methods for dynamic optimisation ............................................................ 13
3.6. Parameter estimation ................................................................................................. 13
4. Results ................................................................................................................................ 14
4.1. Case 1: Monod model ................................................................................................ 14
4.2. Case 2: Extended Monod model................................................................................ 21
4.3. Case 3: Extended Haldane model ............................................................................. 24
5. Discussion .......................................................................................................................... 26
6. Conclusion and recommendations ..................................................................................... 27
7. List of symbols .................................................................................................................... 28
8. References ......................................................................................................................... 29
9. Appendices ......................................................................................................................... 30
9.1. Appendix 1. Derivation of Monod and Lamber-Beer law ........................................... 30
9.2. Appendix 2. Derivation of Haldane and Lamber-Beer law......................................... 31
9.3. Appendix 3. Analytical solution .................................................................................. 33
9.4. Appendix 4. Tables of experimental design............................................................... 36
9.5. Appendix 5. Program scripts...................................................................................... 39
ii
Index of figures
Figure 1. Algae growth model ....................................................................................................... 2
Figure 2. Algae photobioreactor .................................................................................................... 4
Figure 3. Specific growth rate as a function of light intensity for Monod model (a) and Haldane
model (b) ...................................................................................................................... 6
Figure 4. Light attenuation in photobioreactor............................................................................... 7
Figure 5. Piecewise linear optimisation (long dash) and continuous optimisation (line)............. 13
Figure 6. Precision and accuracy ................................................................................................ 13
Figure 7. Optimised input trajectory by using continuous optimisation with (a) q1:q2 = 0:1 which
fall together with q1:q2 = 1:1 and (b) q1:q2 = 1:0......................................................... 15
Figure 8. Sensitivity trajectory and cost function J trajectory with (a) q1:q2 = 0:1 which fall
together with q1:q2 = 1:1and (b) q1:q2 = 1:0 ............................................................... 15
Figure 9. Optimised input trajectory by using piecewise linear with (a) q1:q2 = 0:1E8 which fall
together with q1:q2 = 1E8:1E8 and (b) q1:q2 = 1E8:0 ................................................. 15
Figure 10. Parameter estimation for weighting factor combination without noise for Case 1 ..... 16
Figure 11. Sum of squares contour lines of continuous optimisation with q1:q2 = 1:0 ................ 16
Figure 12. Parameter estimation for weighting factor combination with noise for Case 1 .......... 16
Figure 13. Constant, stepwise, linear increasing, and optimised input of Monod model for 12
hours cultivation ......................................................................................................... 17
Figure 14. Estimated parameters without noise application of Monod model for 12 hours
cultivation ................................................................................................................... 18
Figure 15. Estimated parameters without noise application of Monod model for 12 hours
cultivation ................................................................................................................... 18
Figure 16. Estimated parameters with 1% relative noise application of Monod model for 12
Figure 17. Algae concentration as a result of optimised light intensity for Case 1 ..................... 19
Figure 18. Estimated parameters with 101 data points by using Monod model for 25 hours
cultivation ................................................................................................................... 19
cultivation ................................................................................................................... 20
Figure 20. Optimised input of Monod model of 25 hours cultivation ........................................... 20
Figure 21. Optimised light intensity by using piecewise linear for Case 2 .................................. 21
Figure 22. Optimised input trajectory for Case 2 by using continuous optimisation ................... 22
Figure 23. Algae concentration, sensitivity of parameters, cost function trajectories for Case 2 22
Figure 24. Estimated parameters without noise application of extended Monod model for 12
Figure 25. Estimated parameters with 1% relative noise application of extended Monod model
for 12 hours cultivation ............................................................................................... 23
Figure 26. Input trajectories with weighting factor choices of (a) q1=1E8, q2=0, q3=0; (b) q1=0,
q2=1E8, q3=0; (c) q1=0, q2=0, q3=1E8; (d) dilution rate trajectory for all choices....... 24
Figure 27. Estimated parameters without noise application of extended Haldane model for 12
Figure 28. Estimated parameters with noise application of extended Haldane model for 12 hours
cultivation ................................................................................................................... 25
iii
Index of tables
Table 1. Oil content of some microalgae....................................................................................... 1
Table 2. Artificial light source in cultivation ................................................................................. 14
Table 3. Comparison of weighting factors ................................................................................... 36
Table 4. Light intensity trajectories by using Monod model with 12 hours cultivation................. 36
Table 5. Light intensity trajectories by using Monod model for 25 hours cultivation (noise free) 37
Table 6. Light intensity trajectories by using Monod model and Lambert-Beer law.................... 37
Table 7. Light intensity trajectories by using Haldane model and Lambert-Beer law ................. 38
iv
Acknowledgement
This report is the result of my major thesis work at the Systems and Control Group, Wageningen
University. I am interested in the topic since it is related to algae which have big potential for
renewable energy. I believe that this energy will be widely used in the future. Moreover, the
model use especially Monod and Haldane kinetic model can describe the algae growth
characteristic and it can be analysed through its parameter. This step is an important step to
achieve in this research.
I would like to express my gratitude to all those who gave me the opportunities for doing this
thesis. My sincere thanks to my supervisors Ton van Boxtel and Karel Keesman for the ideas,
suggestions, and enthusiasm during my thesis work and writing this report. Without your support
and guidance, I could not have completed this thesis. Many thanks to Hans Stigter for sharing
his knowledge to solve the analytical solution with mathematica. I also would like to say thanks
to all students who were doing their thesis in the same period with me, especially to Huda who
always shared the information she had, and Rizal, Fazillah and Hassan who are motivating
each other while working in the same room in Technotron building, and to all colleagues in the
Systems and Control Group, who always provided me with a favourable environment for my
work.
Last but not least, I would like to thank to my parent, my wife, and all my Indonesian mates for
their support.
Mochamad Bagus Hermanto
May 2009
v
Summary
Microalgae have high potential to be used for energy production. Growth of algae depends on
inputs as light, temperature, etc. Models are needed to describe growth in algae cultivation. This
thesis aims the design of experimental methods to obtain algae growth kinetics more accurate
than with conventional experimental methods. The growth kinetics that concerned are Monod
and Haldane. In this work, the approach to design experimental procedure is based on
parametric sensitivity function.
Optimal input design method is a way to find a good estimate of parameters by designing
optimal trajectories for the experimental inputs that maximize the parametric sensitivity over an
experiment. Here, optimal input design is applied on three types of model which are Monod,
extended Monod (Monod and Lambert-Beer law), and extended Haldane (Haldane and
Lambert-Beer law). The light attenuation of light intensity inside the photobioreactor can be
described by using Lambert-Beer law equation. As the inputs for the models, light intensity and
dilution rate were selected.
The estimated parameters (max and KI) were obtained for Monod and extended Monod model
by using optimal input design method. Whereas, three parameters (max, K1, and K2) were
estimated for extended Haldane model. The correct choice of weighting factors for this
optimisation was observed to reduce the confidential interval of the estimated parameters.
Then, the estimated parameters were compared with the estimated parameters from intuitive
input trajectories such as constant, stepwise, and linear increasing light intensity for Monod and
extended Monod model. The correlation among the parameters was considered.
By giving weighting factor priority to KI for Monod model and max for Haldane model in
optimisation, the best estimated parameters with low confidential interval were obtained. The
optimised input trajectories in Monod and extended Monod model has the best result in
parameter estimation when it is compared to intuitive input trajectories. This experimental
design can minimise the confidential interval of the estimated parameters, but it did not cancel
the influence of correlation. The noise at the output trajectory will significantly reduce the
precision and the accuracy of the estimated parameter.
The influence of noise to the parameter estimation was observed in order to accommodate the
measurement error in reality. Therefore, it is recommended to add noise filtering in order to
obtain accurate estimate with low confidential interval. Another approach is also recommended
to overcome the correlation problem in parameter estimation.
vi
1. Introduction
The demand of fuel is increasing rapidly faster than the world population growth. The world
population grew by 6.4% from 6.09 billion in 2000 to 6.48 billion in 2005
(http://www.census.gov/ipc/www/idb/worldpop.html), whereby, according to Energy Information
Administration (EIA), the world oil consumption increased by 9.5%, from 76.712 to 84.005
million barrels per day (http://www.eia.doe.gov/pub/international/iealf /table12.xls). Most of the
fuels used nowadays are produced from fossil fuel which is refined and burned. Significant
amounts of energy are distributed among sectors like transportation, residential, commercial,
and industrial. However, fossil fuel is considered as non-renewable energy where the available
quantity has been decreasing.
As renewable energy source, biofuel becomes an alternative source to fulfil the daily needs.
One of the emerging alternatives is biodiesel which substitutes the requirement of fossil-based
for transportation and industries. Biodiesel is derived from vegetable oil which has a long carbon
chain so that the viscosity becomes high. Due to the high viscosity of vegetable oil, incomplete
combustion and carbon decomposition take place when it is applied directly to the engine;
therefore, further processing called transesterification is needed to shorten its carbon chain
structure.
Various oils have been in use in different countries as raw materials for biodiesel production.
Soybean oil is commonly used in United States and rapeseed oil is used in many European
countries, whereas, coconut oil and palm oils are used in Malaysia. Transesterification of edible
oils has also been carried out from the oil of canola and sunflower. Other edible and non-edible
oils, animal fats, algae and waste cooking oils have also been investigated by researchers for
the development of biodiesel (Chisti, 2008). In fact, most of this vegetable oil is produced from
plants used for food production. This development creates a competition between the use for
fuel and food.
Microscopic algae often called as microalgae have fast growth and can produce lipids and have
a high potential to be used for biodiesel production. Microalgae have fast growth due to its
exponential growth. Oil levels of 2050% are quite common (Table1). Oil productivity, that is the
mass of oil produced per unit volume of the microalgae broth per day, depends on the algae
growth rate and the oil content of the biomass. Microalgae with high oil productivities are
desired for producing biodiesel (Chisti, 2007). The total oil and fat content of microalgae ranges
from 1% to 70% of the dry weight and tends to be inversely proportional to the rate of growth
with greater accumulations during stationary phase. The percentage of total lipid as neutral lipid,
glycolipid, and phospholipid also varies widely among and within groups of microalgae
(Guschina and Harwood, 2006).
Table 1. Oil content of some microalgae

Oil content Oil content
Microalgae Microalgae
(% dry wt) (% dry wt)
Botryococcus braunii 2575 Nannochloropsis sp. 3168
Chlorella sp. 2832 Neochloris oleoabundans 3554
Crypthecodinium cohnii 20 Nitzschia sp. 4547
Cylindrotheca sp. 1637 Phaeodactylum tricornutum 2030
Dunaliella primolecta 23 Schizochytrium sp. 5077
Isochrysis sp. 2533 Tetraselmis sueica 1523
Monallanthus salina > 20 Source: (Chisti, 2007)
Nannochloris sp. 2035
1
Some factors influence the algae growth and lipid production. Photosynthetic growth requires
light, carbon dioxide, water and inorganic salts. Temperature must remain generally within 20 to
30 C (Chisti, 2007). To minimize expenses, biodies el production must rely on freely available
sunlight, despite daily and seasonal variations in light levels. Growth medium must provide the
inorganic elements that constitute the algal cell. Essential elements include nitrogen (N),
phosphorus (P), iron and in some cases silicon.
1.1. Problem definition
The interesting part of the algae cultivation is the way to control its productivity by the given
inputs (light, temperature, and other operational conditions). In order to improve and to control
the cultivation, models that describe the cultivation including algae growth and its kinetic
expression are needed (Figure 1).
Algae growth and its kinetic expression are described by using many types of mathematical
equations i.e. Michaelis-Menten or Monod equation (Grima et al., 1994; Holmberg, 1982; Rorrer
and Mullikin, 1999), Haldane equation (Chisti, 2007), and Steele equation (Baquerisse et al.,
1999; Benson et al., 2007).
The design of optimal inputs experiment by employing a model-based optimisation approach to

estimate parameters was becoming a challenge these days. It is supposed that only light
intensity and dilution are the main control inputs and other inputs, such as nutrients and the
temperature remain in the optimal condition. Some experimental designs were conducted in
previous studies. An experimental design which used Monod equation with constant light
intensity and increased dilution rate (A-stat) as the inputs were used in optimisation of
microalgae cultivation parameters (Barbosa et al., 2003). On the other hand, another
experimental design which used constant light intensity and decreased dilution rate (D-stat) as
the inputs was used to determine biomass yield and maintenance coefficient of phototrophic
bacterium (Hoekema et al., 2006).
An alternative experimental design was proposed by using optimal parametric sensitivity control
with Monod model to find the optimal input trajectory. The input is then used to estimate the
parameters in the model (Stigter and Keesman, 2004). This approach for experimental design
should result in the best input design which can produce accurate estimates of cultivation
parameters with low confidential intervals. In this thesis we aim to test the potential for this
method of the estimation of algae growth kinetic.
input (u):
controlled condition
light intensity
state variables (x):

algae concentration (x1)
time
dilution
output (y):
mathematical model algae concentration
time
optimal condition kinetic
expression
input time
temperature
parameter ():
time
specific growth (max)
nutrients saturation constant (KI)
time
Figure 1. Algae growth model
2
1.2. Research objectives
The research aims to evaluate the estimation of kinetic parameters in Monod and Haldane
kinetic for algae growth by using optimal input design based on parametric sensitivities.
1.3. Research approach
In order to accurately estimate the parameter values, the optimal input design approach was
used. The optimal input design aims to find the input trajectory in such a way that the
parameters are optimally estimated with low confidential intervals. Due to this problem, the
optimal parametric sensitivity approach as presented by Stigter and Keesman (2004) was used
as a reference in this thesis. First, Monod type kinetics with affine input is considered. An
optimal input was found based on a singular control arc by applying a parametric sensitivity
function and quadratic cost function using Optimal Control Theory (Bryson, 1999).
Both analytical and numerical solution approaches can be used to find the solution for optimal
input. The analytical approach can be done either by hand or often symbolic manipulation with
Mathematica for simple mathematical models, while, numerical solution is done either using
trajectories with piecewise linear intervals or by continuous trajectories. The trajectories are
obtained from the optimisation of a cost function which contains parametric sensitivity functions,
where parametric sensitivity functions are added as new state variables.
The kinetic expression for growth has an effect on the trajectories for experimental design.
Therefore, two growth kinetic expressions are chosen for that purpose: Monod and Haldane.
For light attenuation, the law of Lambert-Beer is used.
Standard methods by using constant, stepwise, or linear increasing inputs are normally used in
experimental designs. Therefore, in this work of optimal input design, these standard input
strategies are compared with the optimised input in order to estimate parameters. Confidential
intervals of estimated parameters are used in the evaluation of the accuracy of estimation.
1.4. Thesis outline
Chapter 1 defines the problem, research objective, and research approach.

Chapter 2 describes mass balance, models used in this research and its kinetic expression.
Chapter 3 explains optimal input design problem.
Chapter 4 shows the result of some experimental designs.
Chapter 5 discusses the results of the experiment designs.
Chapter 6 concludes the work in this thesis and some recommendations.
3
2. Algae photobioreactors
Photobioreactors have been successfully used for producing large quantities of microalgae
biomass (Chisti, 2007). In addition, photobioreactors are easy to control and permit to have
single-species culture. In algae production, open pond system, flat-plate or flat panel
photobioreactor, and tubular photobioreactor are used. These reactors can be operated as
batch, fed-batch, and continuous bioreactor. In this thesis, a flat-plate photobioreactor system is
considered.
2.1. Photobioreactor model
A mathematical model is a representation of a real system which is usually focused on a set of

selected properties and features of the latter. Models are the essential components for modern
process systems engineering methods (i.e. simulation, optimisation and control), and they are
usually classified into three categories (Banga et al., 2003):
First-principles (or white-box) models, which are derived from well known physical and
chemical relationships, reflecting the underlying principles that govern the process
behaviour.
Data-driven (or black-box) models, which are of empirical nature (e.g. artificial neural
networks, time series).
Hybrid (grey-box) models: a combination of the above.
Generally, in order to develop a model, components involved in a photobioreactor system were

identified and it can be described in Figure 2.
Fin Calg,in
+
Nutrients
Calg

Iin V Iout
Culture chamber
Topt
b = 0.03 m
Fout Calg
Figure 2. Algae photobioreactor
In Figure 2, C A lg [g.l-1] is the algae concentration, Fin or out [l.h-1] the incoming or outgoing
liquid flow in photobioreactor, [ h-1] the specific growth rate of algae, V [l] the photobioreactor
volume capacity, and I [ mol.m
2 1
s ] the light intensity.
4
The mass balance in this system is:
[ mass accumulation] = [ mass flow in] [ mass flow out ] [ mass production or consumption]
d VC A lg
= Fin C A lg,in Fout C A lg + C A lg V (2-1)
dt
dC A lg dV
V + C A lg = Fin C A lg,in Fout C A lg + C A lg V
dt dt
The change of volume is the difference between the incoming and outgoing flow.
dV
= Fin Fout (2-2)
dt
Substituting Equation 2-1, then:
dC A lg
V + C A lg ( Fin Fout ) = Fin C A lg,in Fout C A lg + C A lg V
dt
dC A lg
V = Fin C A lg,in Fout C A lg + C A lgV Fin C A lg + Fout C A lg
dt
Because C A lg,out = C A lg and it assumes that no algae coming into the system ( C A lg,in = 0 ), then:
dC A lg Fin
= C A lg C A lg (2-3)
dt V
In Equation 2-3, the change of algae concentration depends on two terms. The first term is
F
called the dilution factor D = and the second term is the specific growth rate of algae
V
( ) which depends on kinetic expression.
5
2.2. Kinetic expression
The kinetic expression is that part of the model which determines the specific production or
consumption inside a bioreactor system. Michaelis-Menten equation is the most commonly used
in biological, chemical, pharmacological, and medical processes to describe saturation
phenomena. It was first applied to microbiology by Monod (Holmberg, 1982). The equation was
also used as light-limited growth kinetic models with max as maximum specific growth rate, KI
the saturation constant for light intensity, and I the light intensity (Barbosa et al., 2003; Grima et
al., 1994).
I
= max (2-4)
I + KI
For bacterial growth, a non-monotonic kinetic model, Haldane, can be used as well (Versyck et
al., 1997). In the case of light-limited growth kinetic model, it becomes :
I
= max (2-5)
I2
I + K1 +
K2
The parameter K1 indicates how fast the optimum for the specific growth rate max is reached, K2
is the inhibition parameter. The smaller K2 the larger the inhibition effect of the light intensity.
0.1 0.06
0.09
0.05
0.08
0.07
0.04
0.06
0.05 0.03
0.04
0.02
0.03
0.02
0.01
0.01
0 0
0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Light intensity [/mmol/ms] Light intensity [/mmol/ms]
(a) (b)
Figure 3. Specific growth rate as a function of light intensity for Monod model (a) and Haldane
model (b)
Combining Equation 2-3 and Equation 2-4 leads to one state equation.
dC A lg I F
= max C A lg in C A lg (2-6)
dt I + KI V
While by combining Equation 2-3 and Equation 2-5, it becomes:

dC A lg I Fin
= max 2
C A lg C A lg (2-7)
dt I V
I + K1 +
K2
6
2.3. Light attenuation
Light attenuation also contributes to the algae growth limitation due to light transmission and
self shading phenomena. Light intensity is decreasing along path length at photobioreactor
which is presented in Figure 4. Lambert-Beers law was used to determine the average light
intensity and the light gradient inside a flat-plate photobioreactor (Barbosa et al., 2005).
ac .C Al g .b
I out = I in .e (2-8)
CAlg
Iin
Iout
path length (b) = 0.03 m

Figure 4. Light attenuation in photobioreactor
In order to find the average light intensity, integration over the path length inside the reactor
yields:
b
ac .C A lg . x
I in .e dx
I ave = 0
b
1.dx
0
I ave = I in .
1
b
( a .C .b
1 e c A lg .
1
C A lg .ac
) (2-9)
where b is photobioreactor light path and ac is spectral-averaged absorption coefficient on a

dry weight basis (Barbosa et al., 2003). The values of these constants are cited from this
literature and are applied in this work:
b = 0.03 m
2 -1
ac = 200 m kg
7
3. Optimal input design

A good estimate of model parameters can be possibly achieved by applying optimal input
design method (Stigter and Keesman, 2004). Optimal input design method is a way to find a
good estimate of parameters by designing optimal trajectories for the inputs that maximize the
parameter sensitivity over an experiment. In order to maximize the parameter sensitivity, the
system model is extended with additional state variables related to the sensitivity function.
Further, a cost function will be defined to optimize this sensitivity.
3.1. Parametric sensitivity
In parametric models the output sensitivity with respect to a parameter is y/, and
determines whether and how accurate a parameter can be estimated from the input/output data.
In the following it is assumed, without loss of generality, that the states are directly observed so
that y(t)=x(t). If the sensitivity of y with respect to is small or even zero, then the
instrumentation may not be well chosen or the input sequence u(t) is not strong enough to
excite the parametric sensitivities sufficiently (Stigter and Keesman, 2004).
Hence, by using the general dynamical equation:

x& = f ( x, u , , t ) (3-1)
Then, assuming is time invariant, differentiation with respect to and time gives:
f f
x& (t ) x (t ) + (3-2)
x
i.e. the change of parametric sensitivity in time.
Applying these definitions to the algae model and by using x1 for the algae concentration CAlg
yields Equation 3-3 for Monod growth kinetic model or Equation 3-4 for Haldane growth kinetics
model results.
dx1 I F
= max x1 in x1 (3-3)
dt I + KI V
dx1 I Fin
= max 2
x1 x1 (3-4)
dt I V
I + K1 +
K2
The numbers of the sensitivity functions depend on the number of parameters which will be
estimated. In this research, three types of model are considered in three cases:
Case 1 : Monod growth kinetic model
Case 2 : Extended Monod growth kinetic model which is combination of Monod growth
kinetic model and law of Lambert-Beer
Case 3 : Extended Haldane growth kinetic model which is combination of Haldane
growth kinetic model and law of Lambert-Beer
In Case 1, two parameters (max and KI) will be studied. Then, for the additional state variable
with respect to first parameter (KI):
x1
x2 =
K I
8
f 1 f
x&2 = x2 + 1
x1 K I
dx2 F I I
= x2 + max x2 max x (3-5)
KI + I ( KI + I )
2 1
dt V
And for the second additional state variable with respect to second parameter (max):
x1
x3 =
max
f f1
x&3 = 1 x3 +
x1 max
dx3 F I
= x3 + ( x3 max + x1 ) (3-6)
dt V KI + I
In Case 2, when light attenuation in the Monod model is considered, then the first state variable
is given by substituting Equation 2-9 to Equation 3-3.
dx1
I in .
1
b
( a .C .b
1 e c A lg .
1
)
Ca lg .ac Fin
= max x1 x1 (3-7)
1
( )
dt a .C .b 1 V
I in . 1 e c A lg . + K I
b C A lg .ac
The two additional states are derived by using the symbolic toolbox and are given in Appendix
1.
In addition, Haldane growth kinetic model, three parameters (max, K1, and K2) are considered in
Case 3. Therefore, three additional states were added. The first state variable is given by
substituting Equation 2-9 to Equation 3-4.
dx1
I in .
1
b
( a .C .b
1 e c a lg .) 1
C A lg .ac Fin (3-8)
= max 2
x1 x1
dt 1 V
1
( a .C .b
I in . 1 e c A lg .) 1

( )
a .C .b 1 b C A lg .ac
Iin . 1 e c a lg . + K1 +
b C A lg .ac K2
The other three additional states are derived by using the symbolic toolbox and are given in
Appendix 2.
9
3.2. Cost function
The general problem optimises the cost function ( J ) which is defined as:
tf
J = x(t f ) + L( x, u , t ) dt (3-9)
t0
where x(t f ) the terminal condition and L( x, u , p ) the running cost; with start time t0 , final
time t f , and initial condition x(t0 ) specified.
In optimal input design problem, additional state variables of parametric sensitivity need to be
evaluated in time by the running cost which is part of the cost function. Weighting factors are
also added in order to make priorities during optimisation. For the optimisation of the trace of the
Fisher Information Matrix, we define the following quadratic cost function.
tf
J= (q s + q2 s22 + ... + qn sn2 dt )

2
1 1 (3-10)
t0
where qi is the weighting factor, si is the parametric sensitivity, and n is the amount of additional
state variables.
Furthermore, in order to optimize the cost function itself in time, it can be added as one new
additional state variable, which is known as the Mayer formulation.
tf
J = xn +1 = (q s )
+ q2 s22 + ... + qn sn2 dt
2
1 1 (3-11)
t0
If the formulation is expressed into running cost of Case 1 and also at Case 2, then the equation
becomes:
dx4
L= = q1s12 + q2 s22 = q1 x22 + q2 x32 (3-12)
dt
x1
where q1 is the weighting factor for the parametric sensitivity (x2) and q2 is the weighting
K I
factor for sensitivity x1 (x3).
max
3.3. Dynamic optimisation

The continuous dynamic system (Equation 3-1) is described in term of the n-dimensional state
vector x(t) and an m-dimensional input vector u(t) with parameter vector . The optimisation
problem is then to find the control vector u(t) for t0 t tf which minimizes the cost function.
For this purpose Equation 3-1 is adjoined to the cost function (Equation 3-9) with a time varying
Lagrange multiplier vector (t ) (Bryson, 1999):
tf
{ }
J = x(t f ) + L x ( t ) , u ( t ) , t + T ( t ) f x ( t ) , u ( t ) , t x& dt
t0
(3-13)
Define the scalar Hamiltonian function H x ( t ) , u ( t ) , ( t ) , t or H ( t ) for compact notation:
10
H ( t ) = L x ( t ) , u ( t ) , t + T ( t ) f x ( t ) , u ( t ) , t (3-14)
Using Hamiltonian function (Equation 3-14) together with the integration of the term T x& in
Equation 3-13, the equation yields to:
tf
J = x(t f ) T
(t ) x ( t ) + (t ) x ( t ) {H x (t ) , u (t ) , ( t ) , t + & x ( t )} dt
f f
T
0 0
T
(3-15)
t0
Consider an infinitesimal variation in u ( t ) that can be written in term of small changes u ( t ) .

Such variation will produce variations in the state histories x ( t ) and a variation in the
performance index J that could be calculated from (Bryson, 1999):
tf
J = ( x T ) x T x + ( H x + &T ) x + H u u dt (3-16)
t =t f
t =t 0
t0
To avoid having to determine the function x ( t ) produced by u ( t ) , the multiplier T was

chosen in such a way that the coefficients of x ( t ) and x ( t f ) vanish. Bryson makes a few
choices for solving this problem. The first term under the integral should be zero:
H x + &T = 0 , (3-17)
with boundary conditions
T ( t f ) = x ( t f ) . (3-18)
Then, Equation 3-16 develops into:
tf
J = T ( t0 ) x ( t0 ) + H u udt (3-19)
t0
In order to keep J = 0 , H u ( t ) must be zero and either T ( t0 ) or x ( t0 ) . H u ( t ) can be

interpreted as an impulse response function for J . Unit impulse in u at time t1 will
produce J = H u ( t1 ) . Also ( t0 ) J x ( t0 ) , i.e. ( t0 ) is gradient of J with respect to x ( t0 ) ,
&
while holding u ( t ) constant and satisfying Equation 3-1. If x ( t0 ) is specified, then x ( t0 ) = 0 .
For a stationary solution, J = 0 for arbitrary u ( t ) ; this can only happen if:
H u = 0, t0 t t f . (3-20)
Hence, to find a control vector u ( t ) that produces a stationary value of the cost function J , the
optimisation problem must satisfy the following conditions:
State equation
x& = f ( x, u , t ) (3-21)
11
Co-state equation
& = H xT LTx f xT , (3-22)
H uT LTu + T f u = 0
Boundary condition
x ( t0 ) = x0 , (3-23)
(t f ) = T
x (t )
f (3-24)
3.4. Analytical solution

In the same special cases, the solution for optimal input trajectory can be derived analytically,
as is the core when light intensity is considered as the input. To solve the problem light intensity
is considered to be a state variable. Therefore, two state variables are necessary: algae
concentration x1(t) and the integral of the control input x2(t).
max x2 ( t )
f1 = x1 ' ( t ) = x1 ( t )
K I + x2 ( t )
(3-25)
f 2 = x2 ' ( t ) = u ( t ) (3-26)
The Jacobi matrix and the parametric sensitivities are evaluated and augmented to the system:
f1 f1
x2
f x1
=
x f 2 f 2

x1 x2
x1 x1 x2 x2
x3 = , x4 = , x5 = , and x6 =
K I max K I max
Changing the ordering of the derivation with respect to parameters and time is allowed since the
parameter is time invariant according to Equation 3-2. Consequently, we obtain:
f1 f1 f1 f f f1
x x3 + x x5 x1
x4 + 1 x6 1
x2 K I max
x& = S =
1 2
+ (3-27)
f 2 f 2 f 2 f 2 f 2 f 2
x3 + x5 x4 + x6
x1 x2 x1 x2 K I max
By considering only f1 as the function, then it becomes:

f f f f1 f f1
S = 1 x3 + 1 x5 + 1 x4 + 1 x6 + (3-28)
x1 x2 K I x1 x2 max
Thus, the equation above yields:

( x [t ]( KI + x2[t ]) x3[t ] + x1[t ]( x2[t ] + KIx5[t ])) max ( x2[t ]( KI + x2[t ]) x4[t ] + x1[t ]( KIx2[t ] + x2[t ]2 + KI max x6[t ]))
S = max 2
( KI + x2[t ])2 ( KI + x2 [t ])2
The optimisation problem must satisfy the conditions from Equation 3-21 until Equation3-24
which are given in Section 3.6. The complete solution is presented in Appendix 3 by using
Mathematica. The solution is only considered for constant input trajectory. Therefore, the next
sub-chapter constant input is also considered in parameter estimation.
12
3.5. Numerical methods for dynamic optimisation

Bryson developed software to find the optimum trajectories according the conditions given in
Section 3-3. The problem is solved by integration. The state equation is integrated forward in
time, whereas the co-state equation is integrated backward in time. This procedure continuous
H
until the conditions Hu u nearly zero are satisfied. The algorithm delivers for the time

interval under consideration smooth continuous trajectories (see Figure 5 continuous line).
As an alternative for Brysons method which is presented in Section 3.3, optimized trajectories
can be obtained by approximation of the input trajectories by piecewise linear functions over
time intervals (see Figure 5 long dash line). In this case, the optimisation concerns the values at
the edges of each interval. This approach is faster than continuous optimisation since there are
less interval divisions. A large value for the weighting factor is chosen due to convergence rate
improvement. For this approach, fmincon function in Matlab is used.
60
50
40
Input
30
20
10
0
0 1 2 3 4 5 6
Time
Figure 5. Piecewise linear optimisation (long dash) and continuous optimisation (line)
3.6. Parameter estimation

When the optimal trajectory inputs are obtained then it is used in model simulation. The
obtained data with or without noise are fitted to estimate the parameters by an error norm
minimisation. In this step, Matlab function lsqnonlin is used.
A good estimate of parameter should have high precision and high accuracy. An illustration of
precision and accuracy are described in Figure 6. The precision of estimated parameters can be
evaluated by confidential intervals. Therefore, confidential intervals are also calculated by using
nlparci function in Matlab.
In this work, we only focused on having a precise estimate of the parameters by properly
choosing the control input u(t).
Figure 6. Precision and accuracy

(Source: Matlab help files)
13
4. Results
Experimental design is made based on the model availability. In Chapter 3.1, three cases are
defined based on the type of the model. Then, some constants are obtained based on previous
study with Dunaliella tertiolecta as a model microalgae (Barbosa et al., 2003).
max = 0.09 h 1
= 69.86 mol.m .s
2 1
KI
with initial algae concentration ( x1 (0) ) is 1 g .l 1 .
In addition, existing types of equipments which are used in experimental design have limitations
on providing light. Based on Table 2, a light source range from 0 mol/ms to 1500 mol/ms is
chosen for upper boundary in the input trajectory during the simulation.
Table 2. Artificial light source in cultivation

type of light source used in
maximum output
real practice
red LED 2200 mol/ms
red/blue LED 2800 mol/ms
60 tungsten-halogen lamps 1500 mol/ms
Source : Carsten Vejrazka, Kenniseenheid Agrotechniek & Voeding
4.1. Case 1: Monod model

A strategy to estimate two parameters (max and KI) by combining cost functions weighting
factors to set a priority of optimisation is discussed. Then, a comparison of constant, stepwise,
linear increasing and optimized light intensity by using Monod growth kinetic; without and with
noise application are presented. Here, it is assumed that the light path is very thin, so that light
extinction can be neglected.
4.1.1. Choice of weighting factors
Equation 3-12 contains two weighting factors which are q1 the sensitivity of KI and q2 the
sensitivity of max. The optimisation is affected by q1 and q2.
The first step, optimisation result by using q1: q2= 0 : 1; q1: q2 = 1 : 0; and q1: q2= 1 : 1 as the
weighting factors are compared.
Figure 7, for umax =1500 mol/ms, shows that q1: q2= 0 : 1 and q1: q2= 1 : 1 have trajectories
which coincide with the upper boundary of light intensity. Whereas q1: q2 = 1 : 0 give a unique
increasing trajectory for light intensity. For all choices of the weighting factor, the optimised
dilution rate is set equal to zero. This examination shows that q2 has a dominant effect in this
optimisation. The reason is that the sensitivity of max is always much higher than the sensitivity
of KI for all the choices of weighting factors in Figure 8. Therefore, the cost function J value is
dominated by q2 which also becomes high. As a consequence of the dominance is max will be
estimated more accurately than KI. If we want to estimate KI accurately then it is recommended
to give q1 a high value.
14

2000 120
1500 115
1000 110
500 105
0 100
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
1 1
Input Dilution [/h]
0.5 0.5
Dilution [/h]
0 0
-0.5 -0.5
-1 -1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
(a) (b)
Figure 7. Optimised input trajectory by using continuous optimisation with (a) q1:q2 = 0:1 which fall
together with q1:q2 = 1:1 and (b) q1:q2 = 1:0
-3
x 10
Sensitivity KI
0
0
Sensitivity KI
-0.005
-1
-0.01
-2 0 2 4 6 8 10 12
0 2 4 6 8 10 12 Time [h]
Time [h]
Sensitivity mumax
Sensitivity mumax
40 20
20 10
0
0 0 2 4 6 8 10 12
0 2 4 6 8 10 12
Time [h]
Time [h]
-4
4000 x 10
2
2000 1
J
0 0
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
(a) (b)
Figure 8. Sensitivity trajectory and cost function J trajectory with (a) q1:q2 = 0:1 which fall together
with q1:q2 = 1:1and (b) q1:q2 = 1:0
Analogue with continuous optimisation, an optimised trajectory is obtained by using piecewise

linear approach with weighting factor of q1 : q2 = 1E8 : 0 with 12 intervals. A choice of large
value as the weighting factor is chosen due to convergence rate improvement. This approach is
conducted in order to assure the right trajectory from the continuous optimisation.
Figure 9 shows that the light intensity trajectories which are obtained by piecewise linear
optimisation have similar trajectories with those which are obtained by continuous optimisation.
Light Intensity [/mmol /ms]
120
2000
1500 115
1000 110
500 105
0 100
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
(a) (b)
Figure 9. Optimised input trajectory by using piecewise linear with (a) q1:q2 = 0:1E8 which fall
together with q1:q2 = 1E8:1E8 and (b) q1:q2 = 1E8:0
The next step is to examine the results of the parameter estimation. The output trajectories
which are used in this optimisation are obtained from optimised input trajectories with different
choices of weighting factors without noise.
Obtained results are presented in Table 3 (Appendix 4) and Figure 10. By maximising only the
sensitivity of KI with the choice of q1: q2 = 1 : 0, accurate and precise estimation is obtained for
both max and KI.
15
Whereas, by using the other two weighting factors choices, parameter max is almost accurately
estimated, but it is not accurate for KI. In addition, both estimated parameters are not precise
due to a small deviation of the results.
Therefore, the weighting factor choice of q1: q2 = 1 : 0 is recommended in order to have better
estimate of both parameters.
0.0906 80 76.9806
0.0905 0.0904 0.0904 78
76.9806
0.0904 76
0.0903 Mumax high KI high
74
0.0902
0.09 Mumax low 72 KI low
0.0901 69.8595
Mumax 70 KI
0.09
0.0899 68
0.0898 66
0.0897 64
q1:q2 = 0:1 q1:q2 = 1:0 q1:q2 = 1:1 q1:q2 = 0:1 q1:q2 = 1:0 q1:q2 = 1:1
(- noise) (- noise) (- noise) (- noise) (- noise) (- noise)
Figure 10. Parameter estimation for weighting factor combination without noise for Case 1
From Figure 11, the contour lines show that a wide range of parameter combination can be
found due to the strong correlation.
Sum of squares surface Sum of squares contour lines
0.6
0.4
76 0.2
1
0.1
0.1
0.3
74
0.5
0.2
0.8
72
0.6
0.3
0.2
70
KI
0.4
0.1
0.1
0.3
0.2
0.4
0.2 68
0.3
0
0.6
66
0.2
80
0.7
0.1
0. 1
0. 5
70 0.105 64 0.8
0.1
0.2
0.4
0.095
0.09
60 0.085 0.082 0.084 0.086 0.088 0.09 0.092 0.094 0.096 0.098
0.08
KI
/m /m
max max
Figure 11. Sum of squares contour lines of continuous optimisation with q1:q2 = 1:0
As the next step, parameter estimation is conducted by applying 1 % relative random noise at
the output as a consequence of measurement error. Afterwards, the parameters are estimated
using optimised input trajectories with different choices of weighting factors.
As a result, the parameters are not accurately estimated for all the choices of weighting factors.
However, the choice of q1: q2 = 1 : 0 gives the lowest confidential interval among the other
choices. Therefore, this choice of weighting factor is recommended to estimate the parameters
from noisy output trajectory. The result is presented in Figure 12 and the data is presented in
Appendix 4.
200
3E+06
150
2E+06
100
1E+06
50 65.1978 KI high
0.0911 0.0697 0.0898 Mumax high 89.8435 30.2686
0 Mumax low 0E+00 KI low
q1:q2 = 0:1 q1:q2 = 1:0 q1:q2 = 1:1 Mumax q1:q2 = 0:1 q1:q2 = 1:0 q1:q2 = 1:1 KI
-50 -1E+06
(+ noise) (+ noise) (+ noise) (+ noise) (+ noise) (+ noise)
-100
-2E+06
-150
-3E+06
-200
Figure 12. Parameter estimation for weighting factor combination with noise for Case 1
16
4.1.2. Light intensity and dilution rate trajectories during optimisation
It is observed that the optimised dilution rate is zero for all weighting factor choices in Figure 7.
This is a logical outcome if we regard to Equation 3-5 and Equation 3-6. It is observed that the
dilution rate term (D = F / V) has negative contribution to the maximisation of sensitivity of max
and sensitivity of KI.
Whereas, it is shown that a unique trajectory is obtained as optimised light intensity depends on
the weighting factors choice. This is also a logical outcome from Equation 3-5 and Equation 3-6.
It is observed that light intensity (I) has strong contribution to the maximisation of sensitivity of
max and sensitivity of KI.
4.1.3. Comparing optimised trajectories with alternative experimental practice
In this step, the optimised trajectories are compared with trajectories which are used in
experimental practice. Three types of trajectories are considered. The applied values are
derived from the optimised trajectories:
constant light intensities are chosen which are
-2 -1
(i) at lower level of optimisation result (102.5 mol.m s ) and
-2 -1
(ii) at upper level of optimisation result (118 mol.m s ),
stepwise light intensity is chosen at the lower level and the upper level of optimised
trajectory,
linear increasing light intensity is chosen from the lower level until the upper level of
optimised trajectory and fourthly, the optimised trajectory by giving priority to KI.
The effectiveness of these trajectories to estimate the kinetic parameters are compared with the
optimised trajectory for KI. For all these cases, the optimised dilution rate trajectory is zero. The
trajectories are presented in Figure 13.
120
120
115 115
110 110
105 105
100 100
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
a) lower level constant light b) upper level constant light

(constant 1) (constant 2)
120
120
115 115
110 110
105 105
100 100
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
1
c) stepwise light d) linear increasing light
120 1
115
Dilution rate [/h]
0.5
110 0
105 -0.5
100 -1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
e) optimised light f) optimised dilution rate
Figure 13. Constant, stepwise, linear increasing, and optimised input of Monod model for 12 hours
cultivation
17
As the next step, input trajectories are used in simulation for 12 hours cultivation. The total of
101 data points of algae concentration are taken from the output trajectories without additional
noise as measurement data. Afterwards, measured data are compared with the model output
trajectory to extract the parameters. Obtained results are presented in Table 4 (Appendix 4).
Figure 14 shows that parameters for both parameters can be accurately estimated except for
two types of constant light input trajectories.
0.14 200
0.12 150
0.1 0.0900 0.0900 0.0900
Mumax high 100 69.8127
0.0688 69.8370 69.8595
0.08 0.0662
Mumax low
50 24.2849 25.5948 KI high
0.06 Mumax KI low
0.04 0 KI
constant 1 constant 2 stepwise linear optimised
0.02
-50 (- noise) (- noise) (- noise) increasing (- noise)
0 (- noise)
constant 1 constant 2 stepwise linear optimised -100
(- noise) (- noise) (- noise) increasing (- noise)
(- noise) -150
Figure 14. Estimated parameters without noise application of Monod model for 12 hours cultivation
Estimated parameters from stepwise, linear increasing, and optimised input trajectories are
considered in detail in Figure 15.
It is observed that the estimate value of max is precisely and accurately estimated using those
three input trajectories. Whereas, the estimate value of KI is accurately estimated with the
lowest confidential interval by using optimised input trajectory.
Therefore, parameter estimation by using optimised input trajectory by maximising the cost
function value gives the best result under free noise condition.
0.1000 69.8800
69.8595
0.0950 69.8600
0.0900 0.0900 0.0900
0.0900 69.8400 69.8370
Mumax high KI high
0.0850 Mumax low 69.8200 KI low
69.8127
Mumax KI
0.0800 69.8000
0.0750 69.7800
0.0700 69.7600
stepwise linear increasing optim ised stepwise linear increasing optimised
(- nois e) (- noise) (- nois e) (- noise) (- nois e) (- noise)
Figure 15. Estimated parameters without noise application of Monod model for 12 hours cultivation
As the last step of examination, 1 % relative noise is added as perturbation in the output for
three types of light intensity trajectories. Then, total of 101 obtained data points are used as
measured data and compared with the model output trajectory to extract the parameters.
The results are presented in Figure 16. Since no noise filtering in the estimation, none of
parameters are accurately estimated with those three types of input trajectories. Moreover, a
large deviation of the estimated parameters is obtained from the estimation. However, it can be
drawn that the optimised input trajectory can reduce the noise influence and gives the smallest
confidential interval among the other input trajectories.
18
0.35 600
0.3 500
0.25 400
0.2 300
Mumax high KI high
0.15 Mumax low 200 KI low
180.1684
0.145
Mumax KI
0.1087 106.9057
0.1 100
0.078 46.1518
0.05 0
stepwise linear increasing optimised
0 -100 (+ noise) (+ noise) (+ noise)
stepwise linear increasing optimised
-0.05 (+ noise) (+ noise) (+ noise) -200
Figure 16. Estimated parameters with 1% relative noise application of Monod model for 12 hours
cultivation
4.1.4. The effect of additional data points and cultivation time in parameter estimation
By using the optimised input trajectory and given parameters from Barbosa et al., (2003), algae
concentration increase almost twice of initial algae concentration for 12 hours cultivation and
almost five times of initial algae concentration for 25 hours cultivation (Figure 17). In the
following steps, comparison of additional data points and cultivation time in parameter
estimation are conducted. The results are presented in Table 5 (Appendix 4).
5 5
Algae Concentration [g/l]
4 4
3 3
2 2
1 1
0 2 4 6 8 10 12 0 5 10 15 20 25
Time [h] Time [h]
Figure 17. Algae concentration as a result of optimised light intensity for Case 1
Firstly, a comparison of parameter estimation between 12 hours cultivation and 25 hours

cultivation is conducted with the same amount of data points. Three types of light intensity
trajectories are compared in parameter estimation.
The estimated parameters are presented in Figure 18 for 12 hours cultivation and Figure 19 for
25 hours cultivation. As a result, the extension of cultivation time until 25 hours with 101 data
points has nearly similar result with the 12 hours cultivation. The extension of cultivation time
has no influence with the same data points in parameter estimation.
0.1 0.09 0.09 0.09 69.865
0.09 69.86
0.08 69.8536
69.855
0.07 69.8503 69.85
0.06 Mumax high 69.85 KI high
0.05 Mumax low KI low
69.845 KI
0.04 Mumax
0.03 69.84
0.02
69.835
0.01
0 69.83
stepwise linear increasing optimised stepwise linear increasing optimised
cultivation
Secondly, a comparison of different amount data points with the same cultivation time is
conducted. The estimated parameters with additional data points up to 251 points which are
obtained from 25 hours cultivation are presented in Figure 19.
It yields that extra data points for a longer cultivation increases the accuracy and precision of
estimated parameters.
19
69.865
0.1 0.09 0.09 0.09 69.86 69.86
0.09 69.86
0.08
0.07 69.855
0.06 Mumax high 69.8476

KI high
0.05 69.85 KI low
Mumax low
KI
0.04 Mumax
69.845
0.03
0.02
69.84
0.01
0
69.835
stepwise linear increasing optimised stepwise linear increasing optimised
cultivation
In this case, the optimised light input trajectory in Figure 20 has the same form as in Figure 13e
but the values are at higher level. Whereas, the dilution rate trajectory is zero in time.
Input Light intensity [/mmol / ms]
200 1
Input Dilution [/h]

0.5
180
0
160
-0.5
140 -1
0 5 10 15 20 25 0 5 10 15 20 25
Time [h] Time [h]
Figure 20. Optimised input of Monod model of 25 hours cultivation
20
4.2. Case 2: Extended Monod model

A strategy to estimate two parameters (max and KI) for a system with light attenuation due to the
algae and light transmission due to light path is considered now in combination with Monod
growth kinetic. Also here a comparison of constant, stepwise, linear increasing and optimized
light intensity trajectories by using extended Monod growth kinetic; without and with noise
application are presented.
4.2.1. Optimised light intensity trajectory
A 12 linear intervals trajectory is obtained by using piecewise linear approach with weighting
factor priority to KI. It is presented in Figure 21.
In this case, the light intensity trajectory falls together with the upper boundary of light intensity
after 12 hours cultivation. After that, the optimisation generates constant input in the upper level
of light intensity as shown after 12 hours cultivation.
Therefore, it is concluded that a longer period than 12 hours cultivation has a trajectory that
nearly a constant level at the upper boundary. And according to the result from Case 1, an
accurate estimate of parameters is hardly obtained by using the constant light intensity
trajectory input.
1500 1500
1400
1400
1300
1200 1300
1100
1200
1000
900 1100
800
1000
700
600 900
0 2 4 6 8 10 12 0 5 10 15 20 25
Time [h] Time [h]
Figure 21. Optimised light intensity by using piecewise linear for Case 2
4.2.2. Comparing optimised trajectories with alternative experimental practice
As initial step, a continuous optimisation of 12 hours cultivation is obtained with weighting factor
priority to KI. It is presented in Figure 22a for incoming light intensity trajectory and Figure 22b
for average light intensity trajectory along the path light.
It is observed that the demand of incoming optimised light intensity in Case 2 is larger than the
demand of optimised light intensity in Case 1. An examination showed that the average light
intensity along the light path in Case 2 is just about the same with the optimised light intensity
Case 1. In addition, the optimised dilution rate of Case 2 is similar with Case 1.
Moreover, the growth rate trajectory as the function of light intensity increases quickly in the
initial light intensity. Then, it increases slowly as the light intensity increases. The correlation
between the growth rate and the light intensity is given in Figure 22d and e. The graphs show
that the growth rate is nearly constant.
21
Average Light Intensity [/mmol / ms]

Incoming Light Intensity [/mmol / ms]
1500 130
125
120
1000 115
110
105
500 100
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
Average Light Intensity [/mmol / ms]
a) incoming light intensity b) average light intensity along light path

0.06 0.06
0.04 0.04
/m [/h]
/m [/h]
0.02 0.02
0 0
600 700 800 900 1000 1100 1200 1300 1400 1500 105 110 115 120 125 130
Light Intensity [/mmol / ms] Light Intensity [/mmol / ms]
c) growth rate as a function of incoming light intensity d) growth rate as a function of average light intensity
1
0.5
Dilution [/h]
-0.5
-1
0 2 4 6 8 10 12
Time [h]
e) optimised dilution rate
Figure 22. Optimised input trajectory for Case 2 by using continuous optimisation
As a result of given input during cultivation, the algae concentration in Case 2 has the same
amount as in Case 1 (Figure 23). Moreover, even without giving weighting factor at max, it is
observed in Figure 23 that the sensitivity of max has stronger value than the sensitivity of KI.
While the cost function J is maximised in time.
2
Sensitivity KI
0
1.9
-0.005
1.8
-0.01
0 2 4 6 8 10 12
1.7
Time [h]
Sensitivity mumax
1.6
20
1.5
10
1.4
0
0 2 4 6 8 10 12
1.3
Time [h]
-4
1.2 x 10
2
1.1
1
J
1
0 2 4 6 8 10 12 0
0 2 4 6 8 10 12
Time [h]
Time [h]
Figure 23. Algae concentration, sensitivity of parameters, cost function trajectories for Case 2
As the next step, parameters estimation is conducted by using continuous optimisation input
trajectory. Again, the estimated parameters is compared with the estimated parameters which is
obtained by using standard input trajectories which are constant, stepwise, and linear increasing
light intensity. Total of 101 data points from output trajectory free noise are used in this case.
The results are presented in Figure 24 and Table 6 (Appendix 4).
It is observed that all input trajectories give accurate and precise estimate of max. However, the
estimate of KI is accurate for constant, stepwise, and optimised light intensity trajectories. But,
by using linear increasing light intensity trajectory, the estimate of KI is not accurate with quite
22
high confidential interval compared to the others. Significantly, the most accurate estimate of KI
with the lowest confidential interval is obtained by using optimised trajectory.
Therefore, optimised input trajectory is recommended for accurate and precise results in
extended Monod growth kinetic model.
0.1 0.09 0.09 0.09 0.09 0.09 69.8695

69.88 69.8665
0.09 69.8596
69.87 69.8561
0.08
69.86
0.07 69.8356 KI high
Mumax high
69.85
0.06 69.84
0.05 Mumax low KI low
69.83
0.04 Mumax 69.82 KI
0.03 69.81
0.02 69.8
0.01
69.79
0 constant 1 stepwise optimised
constant 1 constant 2 stepwise linear optimised (- noise) (- noise) (- noise)
(- noise) (- noise) (- noise) increasing (- noise)
(- noise)
Figure 24. Estimated parameters without noise application of extended Monod model for 12 hours
cultivation
The last step of examination, additional 1 % relative noise is added as perturbation in the output
for standards light intensity trajectories and optimised light intensity trajectory. Then, total of 101
obtained data points are used as measured data and compared with the model output trajectory
to extract the parameters.
The results are presented in Figure 25. Since no noise filtering in the estimation, none of
parameters are accurately estimated with those input trajectories. Moreover, a large deviation of
the estimated parameters is obtained from the estimation. However, it can be drawn that the
noise influences the accuracy and precision of estimated parameters.
0.12 120
0.1 100
0.0839 0.0835 0.0852
0.0771 0.0738
0.08 80
59.7048 60.836
Mumax high
51.339
0.06 Mumax low 60 42.2548 KI high
Mumax 32.293
40 KI low
0.04
KI
20
0.02
0
0
constant 1 constant 2 stepwise linear optimised
constant 1 constant 2 stepwise linear optimised -20
(+ noise) (+ noise) (+ noise) increasing (+ noise) (+ noise) (+ noise) (+ noise) increasing (+ noise)
(+ noise) -40 (+ noise)
Figure 25. Estimated parameters with 1% relative noise application of extended Monod model for
12 hours cultivation
23
4.3. Case 3: Extended Haldane model

A strategy to estimate three parameters (max, K1, and K2) in the Haldane growth kinetic is
discussed. Then, the evaluation of optimized light intensity trajectories by using Haldane growth
kinetic; without and with noise application are discussed.
4.3.1. Choice of weighting factors

In order to estimate three parameters, three parametric sensitivity equations are needed.
Therefore, three weighting factors are needed which are q1 the sensitivity of K1, q2 the sensitivity
of K2, and q3 the sensitivity of max. The optimisation is affected by q1, q2, and q3.
The trajectory optimisation according to Bryson takes too much iteration to be successful.
Therefore, the optimised input trajectories are obtained using piecewise linear approach. This
optimisation is about the simple version of the continuous optimisation with limited optimisation
points. The optimised light input trajectories have nearly linear increasing trajectories with
different value for every choice of optimisation. It is presented in Figure 26. In all cases, the
optimal dilution trajectory remains zero in time.
110 360
340
100
320
90
300
80 280
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
(a) priority to K1 (b) priority to K2

Light Intensity [/mmol/ms]
220 1
200
Dilution rate [/h]
0.5
180 0
160 -0.5
140 -1
0 2 4 6 8 10 12 0 2 4 6 8 10 12
Time [h] Time [h]
(c) priority to max (d) dilution rate
Figure 26. Input trajectories with weighting factor choices of (a) q1=1E8, q2=0, q3=0; (b) q1=0,
q2=1E8, q3=0; (c) q1=0, q2=0, q3=1E8; (d) dilution rate trajectory for all choices
4.3.2. Parameter estimation using optimised input
For the examination of the parameter estimation the output trajectories which are obtained from
optimised input trajectories with different choices of weighting factors with and without noise
application are applied.
Obtained results are presented in Table 7 (Appendix 4). The sensitivity of max is more dominant
among the other sensitivities. Therefore, it is shown that the choice of q1=0, q2=0, q3=1E8 has
the lowest confidential interval among the other choices for both noise free (Figure 27) and with
noise application (Figure 28).
Due to a strong correlation among three parameters of Haldane kinetic growth, parameter
estimation of these parameters is difficult. Stepwise approach which analyse the parameter one
by one can be a solution. However, it is suggested to apply it after parameter estimation with
three parameters procedure in order to find every parameter with low confidential interval.
Different initial guesses of parameter converge into different estimate of parameters.
24
In free noise case, the larger the sensitivity, the faster it converge to the expected value. The
parameter max has the largest sensitivity value; therefore the value is close to the expected
value. The results with noise seem reasonable, however, the final values (K1; K2) are close to
the initial guest values. So, the identification did not improve the parameter values due to low
sensitivity value of the parameters.
200 25
20
150
69.1871 69.1219 15
69.0258 10.1272 10.0991 10.12 K2 high
100 K1 high
K1 low 10 K2 low
50 K1 K2
5
0 0
q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = q1:q2:q3 =
-50 1E8:0:0 0:1E8:0 0:0:1E8 -5 1E8:0:0 0:1E8:0 0:0:1E8
(- noise) (- noise) (- noise) (- noise) (- noise) (- noise)
0.2
0.18
0.16
0.14
0.12 Mumax high
0.0892 0.0892 0.0891
0.1 Mumax low
0.08 Mumax
0.06
0.04
0.02
0
q1:q2:q3 = q1:q2:q3 = q1:q2:q3 =
1E8:0:0 0:1E8:0 0:0:1E8
(- noise) (- noise) (- noise)
Figure 27. Estimated parameters without noise application of extended Haldane model for 12 hours
cultivation
2E+05 2.50E+04
2E+05 2.00E+04
1.50E+04
1E+05
1.00E+04
5E+04
68.9912 69.0013 69.0001 K1 high 5.00E+03 8.9385 9.0153 9.0003 K2 high
0E+00 K1 low 0.00E+00 K2 low
q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = K1 -5.00E+03 q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = K2
-5E+04
1E8:0:0 0:1E8:0 0:0:1E8 0:0:1E8 0:1E8:0 1E8:0:0
(+ noise) (+ noise) (+ noise)
-1.00E+04
-1E+05 (+ noise) (+ noise) (+ noise)
-1.50E+04
-2E+05 -2.00E+04
-2E+05 -2.50E+04
250
200
150
100
50 0.0914 0.0964 0.094 Mumax high
0 Mumax low
-50 q1:q2:q3 = q1:q2:q3 = q1:q2:q3 = Mumax
1E8:0:0 0:1E8:0 0:0:1E8
-100
(+ noise) (+ noise) (+ noise)
-150
-200
-250
Figure 28. Estimated parameters with noise application of extended Haldane model for 12 hours
cultivation
25
5. Discussion
The analytical approach was used as a starting point and yielded a constant input while the
numerical optimisation methods yielded in time varying trajectories for the input. In this work, an
explanation for the difference was not found. And so a further comparison of the methods is
required.
For noise free experimental design, a comparison with intuitive experimental practice inputs
using Monod model, the best estimate of parameters are obtained by using the optimised input
trajectory. Input strategies based on stepwise input and linear increasing are better than using
constant input in most of cases. By using a range of light intensity values, a range of growth rate
() can be obtained to estimate the specific growth rate (max) and saturation constant for light
(KI). From practice, it was observed that max was best estimated from high level of light intensity
and KI was best estimated from low level of light intensity.
By using extended Monod (Monod + light attenuation) model, constant inputs yielded a better
estimate of parameters than linear increasing light intensity inputs at noise free experimental
design which is out of the expectation (Figure 24). There is a probability that more points were
laid in low level of light intensity so that the parameters especially KI can be well estimated.
The influence of the light intensity to the light path length and algae concentration is conducted.
It is observed that the light intensity demand increases along with the algae concentration and
light path length. The optimum light path length is not considered for reasonable light intensity in
this thesis. Therefore, a consideration of optimum light path length is important because it will
influence the incoming light intensity level.
Despite trajectories for optimal sensitivity have been developed, the estimation results show a
high correlation between the parameters (Figure 11). So, sensitivity optimisation improves the
quality of estimation but does not cancel the correlation.
An accurate and precise estimate of parameters can be obtained by parameter estimation of

Monod growth kinetic model with noise free design. However, if data is subject to noise the
accuracy of the results goes down. As conducted in this research, the noise is located only at
the output trajectory. The parameter estimation is conducted by using noise free input
trajectories model. Therefore, noise filtering is needed in order to minimize the effect of the
noise in parameter estimation.
Three types of model have been presented. For Monod model and extended Monod model, the
best result of parameter estimation is obtained by giving priority to the sensitivity of KI. While
extended Haldane model, the best result of parameter estimation is obtained by giving priority to
the sensitivity of max.
26
6. Conclusion and recommendations

The objective of this result was to evaluate the estimation of kinetic parameters from the Monod
and Haldane kinetic for algae growth by using optimal input design based on parametric
sensitivity. Hereby, three models are used which are Monod model with 2 parameters, extended
Monod (Monod + light attenuation) model with 2 parameters, and extended Haldane (Haldane +
light attenuation) with 3 parameters.
In order to obtain the parameters, analytical and numerical approaches have been conducted.
Analytical approach in this thesis is only possible for systems that are affine in the input,
therefore, constant input trajectory is considered as the solution for this approach. Numerical
approaches are conducted by using either continuous optimisation or piecewise linear
optimisation. Both methods yield similar input trajectories.
General method which is used in this thesis is described as follows:

a. Determine the kinetic expressions together with photobioreactor model
b. Determine parametric sensitivity equations from the model.
c. Obtain optimal input trajectories by analytical solution or dynamical optimisation.
d. Estimate parameters by using optimised input trajectories.
Case 1, Monod model:

a. Two parameters are estimated using this model which are the specific maximum growth
rate max and the light saturation constant KI.
b. The parametric sensitivity of max is larger than KI, therefore, the optimal input trajectories
are depend at the weighting factors for the parameters.
c. Giving priority of weighting factor only for KI is the best choice to obtain optimised input
trajectory for parameter estimation.
d. Additional data from the extension of cultivation length increases the accuracy and precision
of parameter estimation.
Case 2, extended Monod model:

a. Two parameters are estimated using this model which are the specific maximum growth
rate max and the light saturation constant KI.
b. Giving priority of weighting factor only for KI is the best choice to obtain optimised input
c. Light intensity demand increases along with the algae concentration and light path length.
Case 3, extended Haldane model:

a. Three parameters are estimated using this model which are the specific maximum growth
rate max, the light saturation constant K1, and the inhibition constant K2.
b. Giving priority of weighting factor for max is the best choice to obtain optimised input
For those three cases,

a. Despite optimising the sensitivity function, the correlation between these parameters is
strong, therefore, there are a wide range of parameter combination can be found.
b. Initial guess determines the estimated parameter values, indicating the existence of local
minima.
c. Noise on measured data significantly reduces the precision and the accuracy of the
estimated parameters.
In this thesis, three cases are studied in noise free and under noisy conditions. It is observed
from the practice that the estimate of the parameters under noisy conditions is inaccurate and
imprecise. Therefore, one of the recommendations is to add noise filtering to deal with noisy
condition.
In fact, sensitivity optimisation improves the quality of estimation but does not cancel the
correlation. Alternative methods are then needed to obtain more accurate parameter estimation
and minimise the correlation problem.
27
7. List of symbols
notation meaning unit
1
C A lg algae concentration g.l
C A lg,in algae density coming into photo bioreactor g.l 1
d depth of the suspension in flat-plate

photobioreactor
[ m]
h 1
D dilution rate
Fin water flow coming into photo bioreactor m3 .s 1
Fout water flow coming out from photo bioreactor m3 .s 1
H (t ) scalar Hamiltonian function
I light intensity mol.m2 .s 1
I in light intensity coming into photobioreactor mol.m2 .s 1
I ave average light intensity mol.m2 .s 1

J cost function
KI half-saturation constant for light intensity mol.m2 .s 1
K1 half-saturation constant for light intensity mol.m2 .s 1
K2 half-saturation constant for light intensity mol.m2 .s 1

L running cost
n amount of additional state variables
qi weighting factor
si parametric sensitivity
t0 start time s
tf final time s
m3
V photo bioreactor volume capacity
x1 algae concentration g.l 1
Greek Letters
notation meaning unit
attenuation constant for glass [m ]
-1
terminal condition
Lagrange multiplier
max specific growth rate of algae at saturation [h ]
-1
28
8. References
Banga JR, Balsa-Canto E, Moles CG, Alonso AA. 2003. Improving food processing using
modern optimization methods. Trends in Food Science & Technology 14(4):131-144.
Baquerisse D, Nouals S, Isambert A, dos Santos PF, Durand G. 1999. Modelling of a

continuous pilot photobioreactor for microalgae production. Journal of Biotechnology
70(1-3):335-342.
Barbosa MJ, Hoogakker J, Wijffels RH. 2003. Optimisation of cultivation parameters in

photobioreactors for microalgae cultivation using the A-stat technique. Biomolecular
Engineering 20(4-6):115-123.
Barbosa MJ, Zijffers JW, Nisworo A, Vaes W, Schoonhoven Jv, Wijffels RH. 2005.
Optimization of biomass, vitamins, and carotenoid yield on light energy in a flat-panel
reactor using the A-stat technique. Biotechnology and Bioengineering 89(2):233-242.
Benson BC, Gutierrez-Wing MT, Rusch KA. 2007. The development of a mechanistic model to
investigate the impacts of the light dynamics on algal productivity in a Hydraulically
Integrated Serial Turbidostat Algal Reactor (HISTAR). Aquacultural Engineering
36(2):198-211.
Bryson AE. 1999. Dynamic optimization. Menlo Park, CA: Addison Wesley Longman.
Chisti Y. 2007. Biodiesel from microalgae. Biotechnology Advances 25(3):294-306.
Chisti Y. 2008. Biodiesel from microalgae beats bioethanol. Trends in Biotechnology 26(3):126-
131.
Grima EM, Camacho FG, Prez JAS, Sevilla JMF, Fernndez FGA, Gmez AC. 1994. A
mathematical model of microalgal growth in light-limited chemostat culture. Journal of
Chemical Technology & Biotechnology 61(2):167-173.
Guschina IA, Harwood JL. 2006. Lipids and lipid metabolism in eukaryotic algae. Progress in
Lipid Research 45(2):160-186.
Hoekema S, Douma RD, Janssen M, Tramper J, Wijffels RH. 2006. Controlling light-use by
Rhodobacter capsulatus continuous cultures in a flat-panel photobioreactor.
Biotechnology and Bioengineering 95(4):613-626.
Holmberg A. 1982. On the practical identifiability of microbial growth models incorporating

Michaelis-Menten type nonlinearities. Mathematical Biosciences 62(1):23-43.
Rorrer GL, Mullikin RK. 1999. Modeling and simulation of a tubular recycle photobioreactor for
macroalgal cell suspension cultures. Chemical Engineering Science 54(15-16):3153-
3162.
Stigter JD, Keesman KJ. 2004. Optimal parametric sensitivity control of a fed-batch reactor.
Automatica 40(8):1459-1464.
Versyck KJ, Claes JE, Impe JFV. 1997. Practical Identification of Unstructured Growth Kinetic
by Application of Optimal Experimental Design. Biotechnology Progress 13(5):524-531.
29
9. Appendices
9.1. Appendix 1. Derivation of Monod and Lamber-Beer law
dx1dt =
mumax*(Iin*(1/b)*(1-exp(-ac*x1*b))*(1/(x1*ac)))
/(KI+(Iin*(1/b)*(1-exp(-ac*x1*b))*(1/(x1*ac))))*x1
- D*x1
dx2dt =
-(-x2*mumax*Iin*exp(-ac*x1*b)*ac^2*x1^2*b^2*KI
-x2*mumax*Iin^2
+2*x2*mumax*Iin^2*exp(-ac*x1*b)
-x2*mumax*Iin^2*exp(-2*ac*x1*b)
+x2*D*KI^2*ac^2*x1^2*b^2
+2*x2*D*KI*ac*x1*b*Iin
-2*x2*D*KI*ac*x1*b*Iin*exp(-ac*x1*b)
+x2*D*Iin^2
-2*x2*D*Iin^2*exp(-ac*x1*b)
+x2*D*Iin^2*exp(-2*ac*x1*b)
+mumax*Iin*b*ac*x1^2
-mumax*Iin*b*ac*x1^2*exp(-ac*x1*b))
/(KI*ac*x1*b+Iin-Iin*exp(-ac*x1*b))^2
dx3dt =
-(-x3*mumax*Iin*exp(-ac*x1*b)*ac^2*x1^2*b^2*KI
-x3*mumax*Iin^2
+2*x3*mumax*Iin^2*exp(-ac*x1*b)
-x3*mumax*Iin^2*exp(-2*ac*x1*b)
+x3*D*KI^2*ac^2*x1^2*b^2
+2*x3*D*KI*ac*x1*b*Iin
-2*x3*D*KI*ac*x1*b*Iin*exp(-ac*x1*b)
+x3*D*Iin^2-2*x3*D*Iin^2*exp(-ac*x1*b)
+x3*D*Iin^2*exp(-2*ac*x1*b)
-Iin*x1^2*KI*ac*b-Iin^2*x1
+2*Iin^2*x1*exp(-ac*x1*b)
+Iin*x1^2*exp(-ac*x1*b)*KI*ac*b
-Iin^2*x1*exp(-2*ac*x1*b))
/(KI*ac*x1*b+Iin-Iin*exp(-ac*x1*b))^2
30
9.2. Appendix 2. Derivation of Haldane and Lamber-Beer law
dx1dt =
mumax*(Iin*(1/b)*(1-exp(-ac*x1*b))*(1/(x1*ac)))/(KI1+(Iin*(1/b)*(1-exp(-
ac*x1*b))*(1/(x1*ac)))+((Iin*(1/b)*(1-exp(-ac*x1*b))*(1/(x1*ac)))^2/KI2))*x1 - D*x1
dx2dt =
-(x2*D*Iin^4+x2*mumax*Iin^3*exp(-ac*x1*b)*b^2*x1^2*ac^2*KI2-x2*mumax*Iin*exp(-
ac*x1*b)*b^4*x1^4*ac^4*KI2^2*KI1+x2*D*Iin^4*exp(-4*ac*x1*b)-
2*x2*D*KI1*b^3*x1^3*ac^3*KI2^2*Iin*exp(-ac*x1*b)+6*x2*mumax*Iin^3*b*ac*x1*KI2*exp(-
ac*x1*b)+2*x2*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-ac*x1*b)-
6*x2*D*Iin^3*b*x1*ac*KI2*exp(-ac*x1*b)-2*x2*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-ac*x1*b)-
6*x2*mumax*Iin^3*b*ac*x1*KI2*exp(-2*ac*x1*b)-4*x2*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-
ac*x1*b)-2*x2*D*Iin^3*b*x1*ac*KI2*exp(-3*ac*x1*b)+6*x2*D*Iin^3*b*x1*ac*KI2*exp(-2*ac*x1*b)-
x2*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-2*ac*x1*b)-mumax*Iin*b^3*ac^3*x1^4*KI2^2*exp(-
ac*x1*b)-4*x2*D*Iin^4*exp(-ac*x1*b)-4*x2*D*Iin^4*exp(-3*ac*x1*b)+6*x2*D*Iin^4*exp(-
2*ac*x1*b)-2*x2*mumax*Iin^3*exp(-2*ac*x1*b)*b^2*x1^2*ac^2*KI2+x2*mumax*Iin^3*exp(-
3*ac*x1*b)*b^2*x1^2*ac^2*KI2+2*x2*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-
2*ac*x1*b)+2*x2*mumax*Iin^3*b*ac*x1*KI2*exp(-
3*ac*x1*b)+x2*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-
2*ac*x1*b)+2*x2*D*Iin^3*b*x1*ac*KI2+x2*D*Iin^2*b^2*x1^2*ac^2*KI2^2-
x2*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2-
2*x2*mumax*Iin^3*b*ac*x1*KI2+x2*D*KI1^2*b^4*x1^4*ac^4*KI2^2+2*x2*D*KI1*b^3*x1^3*ac^3*
KI2^2*Iin+2*x2*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2+mumax*Iin*b^3*ac^3*x1^4*KI2^2)/(KI1*b^2*x1^
2*ac^2*KI2+Iin*b*x1*ac*KI2-Iin*b*x1*ac*KI2*exp(-ac*x1*b)+Iin^2-2*Iin^2*exp(-
ac*x1*b)+Iin^2*exp(-2*ac*x1*b))^2
dx3dt =
-(x3*D*Iin^4*exp(-4*ac*x1*b)-4*x3*D*Iin^4*exp(-3*ac*x1*b)+6*x3*D*Iin^4*exp(-
2*ac*x1*b)+6*x3*D*Iin^3*b*x1*ac*KI2*exp(-2*ac*x1*b)-2*x3*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-
ac*x1*b)-6*x3*D*Iin^3*b*x1*ac*KI2*exp(-ac*x1*b)-3*mumax*Iin^3*b*ac*x1^2*exp(-2*ac*x1*b)-
4*x3*D*Iin^4*exp(-ac*x1*b)+x3*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-2*ac*x1*b)-
6*x3*mumax*Iin^3*b*ac*x1*KI2*exp(-2*ac*x1*b)+3*mumax*Iin^3*b*ac*x1^2*exp(-ac*x1*b)-
x3*mumax*Iin*exp(-ac*x1*b)*b^4*x1^4*ac^4*KI2^2*KI1+x3*mumax*Iin^3*exp(-
ac*x1*b)*b^2*x1^2*ac^2*KI2+2*x3*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-
ac*x1*b)+6*x3*mumax*Iin^3*b*ac*x1*KI2*exp(-ac*x1*b)-
2*x3*D*KI1*b^3*x1^3*ac^3*KI2^2*Iin*exp(-ac*x1*b)-4*x3*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-
ac*x1*b)-mumax*Iin^3*b*ac*x1^2-2*x3*D*Iin^3*b*x1*ac*KI2*exp(-
3*ac*x1*b)+x3*mumax*Iin^3*exp(-
3*ac*x1*b)*b^2*x1^2*ac^2*KI2+x3*D*Iin^4+2*x3*mumax*Iin^3*b*ac*x1*KI2*exp(-
3*ac*x1*b)+mumax*Iin^3*b*ac*x1^2*exp(-3*ac*x1*b)-2*x3*mumax*Iin^3*exp(-
2*ac*x1*b)*b^2*x1^2*ac^2*KI2+2*x3*D*Iin^3*b*x1*ac*KI2+x3*D*Iin^2*b^2*x1^2*ac^2*KI2^2-
KI2^2*Iin+2*x3*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2+2*x3*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-
2*ac*x1*b)-x3*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-
2*ac*x1*b))/(KI1*b^2*x1^2*ac^2*KI2+Iin*b*x1*ac*KI2-Iin*b*x1*ac*KI2*exp(-ac*x1*b)+Iin^2-
2*Iin^2*exp(-ac*x1*b)+Iin^2*exp(-2*ac*x1*b))^2
dx4dt =
-(-2*x4*D*KI1*b^3*x1^3*ac^3*KI2^2*Iin*exp(-ac*x1*b)-
4*x4*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-ac*x1*b)-2*x4*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-
ac*x1*b)+Iin*b^3*ac^3*x1^4*KI2^2*exp(-ac*x1*b)*KI1+x4*D*Iin^4*exp(-
4*ac*x1*b)+2*Iin^2*b^2*ac^2*x1^3*KI2^2*exp(-ac*x1*b)-4*x4*D*Iin^4*exp(-3*ac*x1*b)-
Iin*b^3*ac^3*x1^4*KI2^2*KI1+3*Iin^3*b*ac*x1^2*KI2*exp(-ac*x1*b)-
6*x4*D*Iin^3*b*x1*ac*KI2*exp(-ac*x1*b)-4*x4*D*Iin^4*exp(-ac*x1*b)+6*x4*D*Iin^4*exp(-
2*ac*x1*b)-x4*mumax*Iin*exp(-ac*x1*b)*b^4*x1^4*ac^4*KI2^2*KI1+x4*mumax*Iin^3*exp(-
ac*x1*b)*b^2*x1^2*ac^2*KI2+2*x4*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-
31
ac*x1*b)+6*x4*mumax*Iin^3*b*ac*x1*KI2*exp(-
ac*x1*b)+x4*D*Iin^4+2*x4*D*Iin^3*b*x1*ac*KI2+x4*D*Iin^2*b^2*x1^2*ac^2*KI2^2-
KI2^2*Iin+2*x4*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2-Iin^2*b^2*ac^2*x1^3*KI2^2-
Iin^3*b*ac*x1^2*KI2+2*x4*D*KI1*b^2*x1^2*ac^2*KI2*Iin^2*exp(-
2*ac*x1*b)+6*x4*D*Iin^3*b*x1*ac*KI2*exp(-2*ac*x1*b)+x4*D*Iin^2*b^2*x1^2*ac^2*KI2^2*exp(-
2*ac*x1*b)-2*x4*D*Iin^3*b*x1*ac*KI2*exp(-3*ac*x1*b)+Iin^3*b*ac*x1^2*KI2*exp(-3*ac*x1*b)-
3*Iin^3*b*ac*x1^2*KI2*exp(-2*ac*x1*b)-Iin^2*b^2*ac^2*x1^3*KI2^2*exp(-2*ac*x1*b)-
2*x4*mumax*Iin^3*exp(-2*ac*x1*b)*b^2*x1^2*ac^2*KI2+x4*mumax*Iin^3*exp(-
3*ac*x1*b)*b^2*x1^2*ac^2*KI2-6*x4*mumax*Iin^3*b*ac*x1*KI2*exp(-2*ac*x1*b)-
x4*mumax*Iin^2*b^2*ac^2*x1^2*KI2^2*exp(-2*ac*x1*b)+2*x4*mumax*Iin^3*b*ac*x1*KI2*exp(-
3*ac*x1*b))/(KI1*b^2*x1^2*ac^2*KI2+Iin*b*x1*ac*KI2-Iin*b*x1*ac*KI2*exp(-ac*x1*b)+Iin^2-
2*Iin^2*exp(-ac*x1*b)+Iin^2*exp(-2*ac*x1*b))^2
32
9.3. Appendix 3. Analytical solution
33
34
35
9.4. Appendix 4. Tables of experimental design
Table 3. Comparison of weighting factors

S K I : S max = q1 : q2
weighting factor 0:1 1:0 1:1
noise 0% 1% 0% 1% 0% 1%
given parameter in generated max 0.090 0.090 0.090 0.090 0.090 0.090
data
KI 69.86 69.86 69.86 69.86 69.86 69.86
max 0.01 0.01 0.01 0.01 0.01 0.01
initial guess parameter
KI 10 10 10 10 10 10
max high 0.0905 137.9518 0.09 0.0982 0.0905 65.9833
confidential interval
max low 0.0903 -137.7696 0.09 0.0412 0.0903 -65.8038
estimated parameter max 0.0904 0.0911 0.09 0.0697 0.0904 0.0898
K I high 78.9146 2.4061E+06 69.8596 85.8919 78.9146 1.1490E+06
K I low 75.0465 -2.4059E+06 69.8594 -25.3547 75.0465 -1.1489E+06
estimated parameter KI 76.9806 89.8435 69.8595 30.2686 76.9806 65.1978
sensitivity value of max 32.1793 14.0105 32.1793
sensitivity value of K I -0.0018 -0.0071 -0.0018
cost function J 2.6544E+03 1.4996E-04 2.6544E+03
Table 4. Light intensity trajectories by using Monod model with 12 hours cultivation
linear linear
constant 1 constant 1 constant 2 constant 2 stepwise stepwise optimised optimised
light intensity increasing increasing
(- noise) (+ noise) (- noise) (+ noise) (- noise) (+ noise) (- noise) (+ noise)
(- noise) (+ noise)
light intensity trajectory constant 1 constant 2 stepwise linear increasing optimised
noise application 0% 1% 0% 1% 0% 1% 0% 1% 0% 1%
light intensity (mol 102.5 and 102.5 and from 102.5 from 102.5 from 102.5 to from 102.5 to
I 102.5 102.5 118 118
/ms) 118 118 to 118 to 118 118 118
time span (h) t 12 12 12 12 12 12 12 12 12 12
given parameter in max 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090
generated data
KI 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86
initial guess max 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01
parameter
KI 10 10 10 10 10 10 10 10 10 10
max high 0.1261 407.7452 0.1244 677.3865 0.0900 0.1535 0.0900 0.293 0.0900 0.1176
max low 0.0064 -407.6126 0.0132 -677.2496 0.0900 0.0638 0.0900 -0.0031 0.0900 0.0383
estimated
parameter max 0.0662 0.0663 0.0688 0.0684 0.0900 0.1087 0.0900 0.145 0.0900 0.078
KI high 138.9074 7.81E+05 141.6475 1.41E+06 69.8465 195.5567 69.8299 475.1789 69.8596 123.4584
KI low -90.3376 -7.81E+05 -90.458 -1.41E+06 69.8275 18.2546 69.7954 -114.8421 69.8594 -31.1548
estimated
parameter KI 24.2849 24.4474 25.5948 24.3112 69.8370 106.9057 69.8127 180.1684 69.8595 46.1518
fitting tolerance 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10 1.E-10
sensitivity value of KI -0.0071 -0.0071 -0.0071 -0.0071 -0.0071 -0.0071 -0.0071 -0.0071 -0.0071 -0.0071
sensitivity value of max 13.5643 13.5643 14.854 14.854 14.1946 14.1946 14.2184 14.2184 14.0105 14.0105
cost function J 1.4979E-04 1.4979E-040.000148980.00014898 1.4981E-041.4981E-04 1.4990E-041.4990E-04 1.4996E-04 1.4996E-04
36
Table 5. Light intensity trajectories by using Monod model for 25 hours cultivation (noise free)
light intensity constant 1 constant 2 stepwise linear increasing optimised
From From From From
light intensity 155.3418 155.3418
I 155.3418 155.3418 194 194 155.3418 155.3418 155.3418 155.3418
(mol /ms) and 194 and 194
to 194 to 194 to 194 to 194
time span (h) t 25 25 25 25 25 25 25 25 25 25
given
parameter in
max 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090 0.090
generated
data
KI 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86
parameter
KI 10 10 10 10 10 10 10 10 10 10
confidential max high 0.0956 0.0662 0.0967 0.0697 0.09 0.09 0.09 0.09 0.09 0.09
interval
max low 0.0368 0.0662 0.0427 0.0697 0.09 0.09 0.09 0.09 0.09 0.09
estimated
parameter max 0.0662 0.0662 0.0697 0.0697 0.09 0.09 0.09 0.09 0.09 0.09
confidential KI high 83.9749 10.3358 89.5741 10.3769 69.8591 69.86 69.8599 69.8492 69.852 69.86
interval
KI low -63.3096 10.3281 -68.826 10.3683 69.8481 69.86 69.8407 69.8461 69.848 69.86
estimated
parameter KI 10.3326 10.3319 10.3741 10.3726 69.8536 69.86 69.8503 69.8476 69.85 69.86
amount of data 101 251 101 251 101 251 101 251 101 251
cost function J 4.700E-03 4.700E-03 4.600E-03 4.600E-03 4.7171E-03 4.7171E-03 4.723E-03 4.723E-03 4.7323E-03 4.7323E-03
Table 6. Light intensity trajectories by using Monod model and Lambert-Beer law
linear linear
light intensity constant 1 constant 1 constant 2 constant 2 stepwise stepwise increasing increasing optimised optimised
(- noise) (+ noise) (- noise) (+ noise) (- noise) (+ noise) (- noise) (+ noise) (- noise) (+ noise)
light intensity trajectory constant 1 constant 2 stepped increased optimised
noise application 0% 1% 0% 1% 0% 1% 0% 1% 0% 1%
light intensity 634 and 634 and from 634 to from 634 to from 634 to from 634 to
I 634 634 1484 1484
(/m ms) 1484 1484 1484 1484 1484 1484
time span (h) t 12 12 12 12 12 12 12 12 12 12
given parameter in max 0.090 0.090 0.090 0.090 0.09 0.09 0.09 0.09 0.09 0.09
generated data
KI 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86 69.86
parameter
KI 10 10 10 10 10 10 10 10 10 10
max high 0.09 0.0991 0.09 0.089 0.09 0.0939 0.09 0.1095 0.09 0.0976
max low 0.09 0.0687 0.09 0.078 0.09 0.0765 0.09 0.0446 0.09 0.0501
estimated
parameter max 0.09 0.0839 0.09 0.0835 0.09 0.0852 0.09 0.0771 0.09 0.0738
KI high 69.8684 85.6335 69.8724 66.5656 69.8582 78.4786 69.852 110.8656 69.8597 85.0723
KI low 69.8646 33.7761 69.8667 36.1124 69.854 43.1935 69.8192 -26.356 69.8595 -9.8863
estimated
parameter KI 69.8665 59.7048 69.8695 51.339 69.8561 60.836 69.8356 42.2548 69.8596 32.293
cost function J 1.1670E-04 1.1670E-041.0771E-041.0771E-04 1.1973E-04 1.1973E-041.2232E-04 1.2232E-04 1.2287E-04 1.2287E-04
37
Table 7. Light intensity trajectories by using Haldane model and Lambert-Beer law
S K1 : S K2 : S Mumax = q1 : q2 : q3
weighting factors 1E8:0:0 0:1E8:0 0:0:1E8

noise without with without with without with
time span (h) t 12 12 12 12 12 12
noise % 0% 1% 0% 1% 0% 1%
max 0.09 0.09 0.09 0.09 0.09 0.09
given parameter in
generated data K1 69.86 69.86 69.86 0 0 0
K2 10 10 10 0 0 0
max 0.08 0.08 0.08 0.08 0.08 0.08
initial guess parameter K1 69 69 69 69 69 69
K2 9 9 9 9 9 9
max high 0.1126 39.5042 0.1757 184.3894 0.0923 7.3244
max low 0.0658 -39.3214 0.0028 -184.1966 0.0859 -7.1364
estimated parameter max 0.0892 0.0914 0.0892 0.0964 0.0891 0.094
K1 high 89.2152 3.30E+04 160.2763 1.80E+05 72.0081 6.46E+03
K1 low 49.159 -3.28E+04 -22.0325 -1.79E+05 66.0435 -6.32E+03
estimated parameter K1 69.1871 68.9912 69.1219 69.0013 69.0258 69.0001
K2 high 13.6967 5.04E+03 20.8777 1.88E+04 10.5481 8.12E+02
K2 low 6.5577 -5.02E+03 -0.6794 -1.88E+04 9.6919 -7.94E+02
estimated parameter K2 10.1272 8.9385 10.0991 9.0153 10.12 9.0003
fitting tolerance 1.E-05 1.E-05 1.E-05 1.E-05 1.E-05 1.E-05
sensitivity value of K1 -0.0016 -0.0016-0.00047575-0.00047575 -0.0011 -0.0011

sensitivity value of K2 0.0034 0.0034 0.0118 0.0118 0.0092 0.0092
sensitivity value of max 1.8637 1.8637 1.9477 1.9477 2.2746 2.2746
cost function J 9.4577E+02 9.4577E+02 5.1264E+04 5.1264E+041.8985E+091.8985E+09
38
9.5. Appendix 5. Program scripts
The scripts are compiled in a CD which is included in this thesis. It has some files as follows:
General files
o clip : a function for clipping input arrays
o derivation : derive parametric sensitivity equations
o fop0 : a function optimisation by using Bryson method that perform some
iterations
o fop0_b : subroutine of fop0 that perform backward integration
o fop0_f : subroutine of fop0 that perform forward integration
o linvfu_vec : differentiate a vector function of x, u, and theta with respect to u
o linvfx_vec : differentiate a vector function of x, u, and theta with respect to x
o mysurface : provide mesh and contour plot with 10% interval
Monod model files

o fitfunA : calculate error between model and measurement values based on
Bryson approach with Monod model
o fitfunPW : calculate error between model and measurement values based on
piecewise linear approach with Monod and extended Monod model
o fittingA : estimate parameter max and KI based on Bryson approach with
Monod model
o fittingPW : estimate parameter max and KI based on piecewise linear approach
with Monod and extended Monod model
o functionA : a function which has calculation switch of cost function to determine
forward and backward integration by using Monod model
o functionPW_A : a function contains cost function calculation for piecewise linear
approach by using Monod model
o functionPW F_A: a function which calculate state variables and input values from
optimised input values for piecewise linear approach by using Monod
model
o modelA : Monod model for Bryson approach
o modelPW_A : Monod model for piecewise linear approach
o sensitivityA : perform input optimisation with Bryson approach by using Monod
model
o sensPW_A : perform input optimisation with piecewise linear approach by using
Monod model
o testmodelA : perform simulation with constant, stepwise and linear increasing input
trajectories by using extended Monod model
Extended Monod model files

o fitfunB : estimate parameter max and KI based on Bryson approach with
extended Monod model
o fitfunPW : calculate error between model and measurement values based on
piecewise linear approach with Monod and extended Monod model
o fittingB : estimate parameter max and KI based on Bryson approach with
o fittingPW : estimate parameter max and KI based on piecewise linear approach
with Monod and extended Monod model
o functionB : a function which has calculation switch of cost function to determine
forward and backward integration by using extended Monod model
o functionPW_B : a function contains cost function calculation for piecewise linear
approach by using extended Monod model
o functionPW F_B: a function which calculate state variables and input values from
optimised input values for piecewise linear approach by using
o modelB : extended Monod model for Bryson approach
39
o modelPW_B : extended Monod model for piecewise linear approach

o sensitivityB : perform input optimisation with Bryson approach by using extended
Monod model
o sensPW_B : perform input optimisation with piecewise linear approach by using
o testmodelB : perform simulation with constant, stepwise and linear increasing input
trajectories by using extended Monod model
Extended Haldane model files

o fitfunC : estimate parameter max , K1 and K2 based on Bryson approach with
extended Haldane model
o fitfunPW_C : calculate error between model and measurement values based on
piecewise linear approach extended Haldane model
o fitfunPW_C1P : calculate error between model and measurement values based on
piecewise linear approach with extended Haldane model
o fitfunPW_C2P : calculate error between model and measurement values based on
o fittingC : estimate parameter max , K1 and K2 based on Bryson approach with
o fittingPW_C : estimate parameter max , K1 and K2 based on piecewise linear
approach with extended Haldane model
o fittingPW_C1P : estimate parameter one of max or K1 or K2 based on piecewise linear
o fittingPW_C2P : estimate parameter two combination of parameters based on
o fittingPW_C_ID : estimate parameter max , K1 and K2 based on piecewise linear
o functionC : a function which has calculation switch of cost function to determine
forward and backward integration by using extended Haldane model
o functionPW_C : a function contains cost function calculation for piecewise linear
approach by using extended Haldane model
o functionPW_C_ID: a function contains cost function calculation for piecewise linear
approach by using extended Haldane model for 2 inputs optimisation
o functionPW F_C: a function which calculate state variables and input values from
o functionPW F_C_ID: a function which calculate state variables and input values from
extended Haldane model for 2 inputs optimisation
o modelC : extended Haldane model for Bryson approach
o modelPW_C : extended Haldane model for piecewise linear approach
o modelPW_C1P : extended Haldane model for piecewise linear approach for estimating
1 parameter only
o modelPW_C2P : extended Haldane model for piecewise linear approach for estimating
2 parameters
o modelPW_C_ID: extended Haldane model for piecewise linear approach for 2 inputs
optimisation
o sensitivityC : perform input optimisation with Bryson approach by using extended
Haldane model
o sensPW_C : perform light intensity input optimisation with piecewise linear
approach by using extended Haldane model
o sensPW_C_ID : perform light intensity and dilution inputs optimisation with piecewise
linear approach by using extended Haldane model
o testmodelC : perform simulation with constant, stepwise and linear increasing input
trajectories by using extended Haldane model
40

Identiication of Algae Growth Kinetics

Uploaded by

Copyright:

Available Formats

Identiication of Algae Growth Kinetics

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Identiication of Algae Growth Kinetics

Uploaded by

Copyright:

Available Formats

Systems and Control Group

Thesis Systems and Control

0.04 Light intensity [/mmol/ms]

Name course : Thesis project Systems and Control

Student : Mochamad Bagus Hermanto

Supervisors : dr.ir. A.J.B. Boxtel

Mochamad Bagus Hermanto

Table 1. Oil content of some microalgae

1.1. Problem definition

The design of optimal inputs experiment by employing a model-based optimisation approach to

state variables (x):

Figure 1. Algae growth model

1.2. Research objectives

1.3. Research approach

1.4. Thesis outline

Chapter 1 defines the problem, research objective, and research approach.

2.1. Photobioreactor model

A mathematical model is a representation of a real system which is usually focused on a set of

Generally, in order to develop a model, components involved in a photobioreactor system were

Figure 2. Algae photobioreactor

The mass balance in this system is:

2.2. Kinetic expression

Specific growth rate [/h]

While by combining Equation 2-3 and Equation 2-5, it becomes:

2.3. Light attenuation

path length (b) = 0.03 m

where b is photobioreactor light path and ac is spectral-averaged absorption coefficient on a

3. Optimal input design

3.1. Parametric sensitivity

Hence, by using the general dynamical equation:

3.2. Cost function

J= (q s + q2 s22 + ... + qn sn2 dt )

3.3. Dynamic optimisation

Define the scalar Hamiltonian function H x ( t ) , u ( t ) , ( t ) , t or H ( t ) for compact notation:

Consider an infinitesimal variation in u ( t ) that can be written in term of small changes u ( t ) .

To avoid having to determine the function x ( t ) produced by u ( t ) , the multiplier T was

with boundary conditions

Then, Equation 3-16 develops into:

In order to keep J = 0 , H u ( t ) must be zero and either T ( t0 ) or x ( t0 ) . H u ( t ) can be

3.4. Analytical solution

By considering only f1 as the function, then it becomes:

Thus, the equation above yields:

3.5. Numerical methods for dynamic optimisation

3.6. Parameter estimation

Figure 6. Precision and accuracy

Table 2. Artificial light source in cultivation

4.1. Case 1: Monod model

4.1.1. Choice of weighting factors

Light intensity [/mmol / ms]

Light intensity [/mmol / ms]

Analogue with continuous optimisation, an optimised trajectory is obtained by using piecewise

4.1.2. Light intensity and dilution rate trajectories during optimisation

4.1.3. Comparing optimised trajectories with alternative experimental practice

Light intensity [/mmol/ms]

a) lower level constant light b) upper level constant light

e) optimised light f) optimised dilution rate

Firstly, a comparison of parameter estimation between 12 hours cultivation and 25 hours

0.1 0.09 0.09 0.09 69.865

0.06 Mumax high 69.8476