Swarm-Assisted Investment Planning of a Bioethanol Plant

Abstract

Bioethanol is a liquid fuel for which a significant increase in the share of energy sources has been observed in the economies of many countries. The most significant factor in popularizing bioethanol is the profitability of investments in construction of facilities producing this energy source, as well as the profitability of its supply chain. With the market filled with a large amount of equipment used in the bioethanol production process, it is often difficult to make an optimal decision regarding the investment. Another issue is the location of the plant itself. Economic benefits are strongly associated with costs of equipment and materials, the amount of revenue from sales, and transportation costs. This article presents an attempt to solve this problem by using several swarm algorithms new and fast-growing optimisation techniques. By employing ant colony optimization, river formation dynamics, particle swarm optimization, and cuckoo search algorithms in the task of bioethanol plant investment planning, the overall suitability of this type of technique has been tested. Moreover, the results allow us to determine which of the preceding algorithms is the most efficient in the given task.

Full text

Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Introduction
Bioethanol is a liquid fuel obtained from the process
of alcoholic fermentation of biomass or biodegradable
municipal waste (e.g., paper or wood). Year after year
a signicant increase in the share of this type of energy
source is observed in the economies of many countries.
Bioethanol is mainly used in transport as a fuel mixture
with gasoline at concentrations of 10%, 20%, and 85%
(labelled as E10, E20, and E85, respectively) [1-2]. This
solution allows for a signicant reduction of greenhouse
gases released into the atmosphere during combustion of
fuel in car engines [3].
Bioethanol is divided into three types based on the
material used for its production [4-5]:
Pol. J. Environ. Stud. Vol. 26, No. 3 (2017), 1203-1214
*e-mail: grzegorz.redlarski@pg.gda.pl
DOI: 10.15244/pjoes/68151
Original Research
Swarm-Assisted Investment Planning
of a Bioethanol Plant
Grzegorz Redlarski1*, Marek Krawczuk1, Adam Kupczyk2, Janusz Piechocki3,
Dominik Ambroziak1, Aleksander Palkowski1
1Department of Mechatronics and High-Voltage Engineering, Gdansk University of Technology,
G. Narutowicza 11/12, 80-233 Gdansk, Poland
2Department of Production Management and Engineering, Warsaw University of Life Sciences,
Nowoursynowska 164, 02-787 Warsaw, Poland
3Department of Electrical Engineering, Power Engineering, Electronics, and Control Engineering,
University of Warmia and Mazury, M. Oczapowskiego 11, 10-719 Olsztyn, Poland
Received: 7 November 2016
Accepted: 29 December 2016
Abstract
Bioethanol is a liquid fuel for which a signicant increase in the share of energy sources has been
observed in the economies of many countries. The most signicant factor in popularizing bioethanol is
the protability of investments in construction of facilities producing this energy source, as well as the
protability of its supply chain. With the market lled with a large amount of equipment used in the
bioethanol production process, it is often difcult to make an optimal decision regarding the investment.
Another issue is the location of the plant itself. Economic benets are strongly associated with costs
of equipment and materials, the amount of revenue from sales, and transportation costs. This article presents
an attempt to solve this problem by using several swarm algorithms – new and fast-growing optimisation
techniques. By employing ant colony optimization, river formation dynamics, particle swarm optimization,
and cuckoo search algorithms in the task of bioethanol plant investment planning, the overall suitability
of this type of technique has been tested. Moreover, the results allow us to determine which of the preceding
algorithms is the most efcient in the given task.
Keywords: bioethanol, combinatorial problem, investment planning, optimization, swarm algorithms

1204
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
• First generation: sugar beets, wheat, corn, sugar cane,
etc.
• Second generation: waste, residues, and lignocellulose
biomass
• Third generation: algae and seaweed
Only the rst generation of bioethanol is currently
produced and used on a large scale [4]. The second
generation of bioethanol is being implemented for
widespread use in several years, while the third generation
is still under development.
Bioethanol has gained in popularity in recent years,
especially in European countries. This gain is related
to the current policy of the European Union regarding
renewable sources of energy, as well as to the eco-
efciency of bioethanol-based fuels [6]. Increased demand
for bioethanol and other renewable energy sources is due
to several international regulations, which pays increasing
attention on the growing share of renewable energy and
the environmental impact of technology [7-11].
However, a major factor in popularizing bioethanol is
the protability of investing in the construction of facilities
producing this energy source, as well as the protability of
its supply chain. A number of factors affect the price of
bioethanol: the current economic and political situation of
a country, the size of taxes and subsidies, current demand
for biofuels, and production and operating expenses [4,
12]. Therefore, in the initial phases of an investment,
the selection of appropriate equipment, technology of
production, and the location of the plant itself is crucial
[12].
Currently, the market is lled with a large number of
producers of equipment used in bioethanol production.
The equipment varies not only by price, but also by total
capacity, power, and efciency. The multitude of choices
of equipment and a wide range of power, performance,
and capacity make the issue of bioethanol production
line design non-trivial. During the design, one should
rst determine the required minimum capacity of devices
based on expected annual production of bioethanol.
By using devices with high energy efciencies, a direct
reduction in annual emission of harmful substances into
the atmosphere is noticed. This is also crucial for annual
expenditure on energy consumption. In addition, taking
into account a limited budget, the role of the designer of
a production line is highly responsible, and any errors are
associated with prolonged payback time.
The literature mostly presents examples of bioethanol
or biogas production optimisation. Most of the works
concentrate on optimising variables associated with
chemical processes that occur during bioethanol or biogas
production. There are diverse examples of nding an
optimal operating point that satises various constraints,
determining an optimal control strategy, or nding a
constant substrate mixture – all of which lead to optimal
and stable operation of the plant. Examples of such work
follow. Optimising an anaerobic sequencing batch reactor
with the use of articial neural networks and genetic
algorithms demonstrated a clear improvement in biogas
production [13]. Using Particle Swarm Optimization
in optimising substrate feed mix resulted in a 20%
improvement in biogas production [14]. Another use
of particle swarm optimization for optimising values
of certain biogas production process variables (e.g.,
temperature, pH value) in a multi-layer perceptron neural
network model resulted in a 20.8% increase in biogas
production [15].
Another aspect that should be included in the decision
process regarding bioethanol plant investments is plant
location. Economic benets are strongly associated with
costs of materials, amount of revenue from sales, and
transportation costs. In general, production costs of biofuel
are associated with the facility size and location [16].
Therefore, the location of the plant should be optimised to
maximise prot. Celli et al. presented a system based on
genetic algorithms, which enable optimal biomass power
plant distribution [17]. Another example used a mixed-
integer linear programming model to optimise supply and
delivery of ethanol [18]. Mixed-integer linear programming
was also used to optimise the design and planning of
biomass-based fuel supply networks according to nancial
criteria [19]. A mathematical model for optimising cost of
a switchgrass-based biofuel supply chain was developed
likewise using mixed integer linear programming [20].
Biorenery location was based on the transportation
cost of biomass and biofuel. An example of performance
optimisation of a biofuel supply chain was carried out by
a two-stage stochastic mixed-integer linear programming
model with the sample average approximation algorithm
[21]. There are several other examples of mixed-integer
linear programming optimisation of biofuel supply chains
[22-24], which appears to be the most used method at
present.
A limited number of sources present an actual system
for biogas or bioethanol plant design assistance. One is
a piece of software developed by M. Samer to plan and
design biogas plants and their concrete structures [25].
However, despite its capability to specify many details
of the plant (such as dimensions of tanks or the amount
of construction materials), it is incapable of performing
any higher-level optimisation. Optimised results from a
mixed integer non-linear programming model were used
to incorporate structural enhancements in distillation
columns and heat integration inside a bioethanol plant
to reduce steam consumption [26]. Problems regarding
capacity expansions of production and storage facilities of
supply network over time, along with associated planning
decisions, were successfully solved using the sample
average approximation algorithm [27]. There is also a
number of articles that present examples of optimisation
of biofuel production processes by selecting appropriate
technologies used in a renery [28].
As is apparent from the foregoing description, some
of the work uses natural-inspired algorithms, especially
swarm algorithms (particle swarm optimization in
particular) to solve relevant problems. Swarm algorithms
are a set of stochastic metaheuristics used in various
optimisation tasks. An inspiration for their development
was the behaviour of social animals and various natural

1205
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Swarm-Assisted Investment Planning...
phenomena. Their effectiveness in solving complicated,
multidimensional problems has been extensively proven,
and suggests that this will also be the case for the presented
problem. In fact, the effectiveness of swarm algorithms in
optimising biofuel plant supply chains has been proven
[29].
Almost all of the foregoing examples of literature
describe optimisation of biofuel supply chains and
their major constituents, such as feedstock availability,
harvesting capacity, and site locations, transportation
network, storage, or regional economic structure, and
policy. A more technical side of bioethanol plant planning
seems to be omitted, i.e., selection of particular pieces of
equipment that constitute a plant.
This article tries to answer two questions: is it possible
to develop a versatile tool for bioethanol investment
planning, and whether and what swarm algorithms
are effective in solving the presented problem. That
is the problem of planning rst-generation bioethanol
plant construction with minimisation of equipment and
operating costs. Therefore, an optimisation system based
on selected swarm algorithms has been developed. The
following swarm algorithms were tested: ant colony
optimization (ACO), river formation dynamics (RFD),
particle swarm optimization (PSO), and cuckoo search
(CS). The system was developed with its simplicity and
exibility in mind, so that simultaneous optimisation of
plant equipment and location could be done by a single
algorithm. In this respect, the article presents a new
approach to the described problem.
The obtained results allow us to conclude that this
type of methodology can nd a use in making investment
decisions for investors and developers.
Overview of the Bioethanol Plant
Model Used
This article incorporates as its model a rst-generation
bioethanol plant. The rst-generation bioethanol
production process consists of a number of different
minor processes such as washing, cutting, extracting,
carbonation, alcoholic fermentation, and distillation.
Fig. 1 presents in detail the process of bioethanol pro-
duction from sugar beet to the nal product: bioethanol
[10, 30]. The model presented is a simplied one that
considers only the main production process. Whether the
addition of more factors (e.g., power or water processing)
would signicantly affect the model remains an open
question.
The sugar beet production process stands as follows.
Transported in autumn and before further treatment, sugar
beets rst must be washed thoroughly. Barrel washers
clean the beets and separate them from stones, sand, and
weeds. Then sugar beets are cut into what is called slice.
Slice is ooded with suitably prepared warm water to
extract sucrose from the beets. As a result of extraction,
a raw juice containing 13-14.5% sugar is obtained. In
order to purify the raw juice from pollutants, lime milk
is added and then the mixture undergoes the process of
carbonation. After ltering off impurities from the raw
juice, a thin juice is obtained (approximately 15% sugar
content). The next step in the production of bioethanol
is thickening the thin juice to increase its sugar content.
Thin juice is repeatedly passed through evaporators.
At the end of the evaporation process a thick juice,
containing 68-71% sugar, is obtained. Then, using yeast
in the fermentation process, carbon dioxide and ethanol is
produced (5-7% concentration of pure ethanol). The nal
stage of the production of bioethanol is its distillation to
increase the concentration of pure alcohol in the nished
product. Distillation is repeated two or three times
(95-96% ethyl alcohol concentration).
As is apparent from the foregoing description, the
complexity of the production process of bioethanol enforces
the use of a large amount of specialized equipment, such
as: barrel washers, cutting machines, carbonators, lters,
centrifuges, mixing tanks, evaporators, and distillers.
All mentioned equipment, in addition to its costs, has
characteristics that highly affect the production process
and its efciency (such as capacity or energy efciency).
Depending on the expected results, the selection of
particular pieces of equipment is not an easy task.
Methods
Problem Formulation
The problem presented in the introduction requires
the selection of successive elements of the technological
process of bioethanol production. As shown in Fig. 1, the
production process can be separated into different stages
in which particular pieces of equipment sequentially
process the beet juice. For each stage, a list of potential
instruments and their characteristics (price, power,
efciency) can be created. Furthermore, the plant must be
Fig. 1. Considered bioethanol production process.

1206
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
placed in a suitable location, on which all raw material
and transportation costs are dependent (after selecting
appropriate buyers and sellers).
Therefore, a combinatorial problem with one objective
function can be formulated. The objective function is to
minimise expenditures calculated on the basis of initial
plant equipment costs, as well as operation costs and
sum of revenue throughout the rst year. The presented
problem can be written in the following form:
(1)
…where F(eq,s,b,l) is the objective function being the
total costs C incurred by the rst year (2) dependent on
chosen equipment eq, plant location l, sugar beet sellers
s, and bioethanol buyers b. The limitation h(eq,s,b) is the
annual production of bioethanol, which should be greater
or equal to intended minimum production pa. Additionally,
all chosen equipment must form a complete bioethanol
production line Lp, comprised of p production stages (as
in Fig. 1).
(2)
The total costs are dependent, among other things, on
purchase costs for equipment (3) and energy expenses
(4) – all of which depend on characteristics of particular
chosen pieces of equipment.
(3)
(4)
…where:
(5)
Therefore, to support the given equations, a list of
equipment for each type used in the production process
(i.e., barrel washers, cutting machines, carbonators, lters,
centrifuges, mixing tanks, evaporators, and distillers) has
been established (Appendix a)). Their price, efciency,
and capacity were loosely based on existing equipment,
and assigned randomly for purposes of the tests. These
characteristics stand for the basis for determining the value
of objective functions. During the investment simulation it
is crucial to not only choose a particular machine, but to
choose their number as well – as it dictates whether overall
device capacity ts the expected annual production.
In addition, the plant is designed to provide a certain
amount of bioethanol per year. This imposes the use of
constraints to lter out incorrect solutions. This is used
mainly to verify whether all tanks provide adequate
capacity, and whether all beet sellers can provide the
appropriate amount of raw material.
The second part of the presented issue is plant location
optimisation. In contrast to a more sophisticated plant
location model developed by Zhang et al. [20], a simple
model for the purpose of location optimisation has been
prepared. Fig. 2 presents a map of potential locations
for the bioethanol plant. In addition to the free elds for
the plant, the map includes locations of eight sugar beet
suppliers (marked grey) and 10 buyers of fuel (marked
black). The choice of a particular eld is dictated by the
cost of purchase (6) and transportation (8) of sugar beets,
and selling (7) and transport (9) costs of bioethanol. All
sugar beet providers and fuel stations have been assigned
with appropriate purchase prices, randomly distributed
to simulate diverse yet levelled economic circumstances
(Appendix b)). Sugar beets are purchased once a year from
chosen sources, whereas bioethanol is sold on a monthly
basis. Truck tonnage capacity and tank litre capacity
were assumed as constant and equal to, accordingly,
12 t and 28,300 l. Transportation costs are dependent on
the distance between a buyer/seller and the plant, which
was assumed as a Cartesian distance between particular
elds with constant transportation cost per unit ($20 US
in this case). All sellers and buyers have their minimum
and maximum supply and demand, which provides
information as to whether they are suitable for the given
case.
(6)
(7)
(8)
(9)
It is apparent that the presented issue is a combination
of a simplied supply chain design with additional
production optimisation variables. Like scheduling, the
described problem focuses on deciding how to commit
available resources between a variety of possible tasks.
In this manner each of the listed equipment must be
assigned to proper production stages. Each of the listed
sugar beet suppliers and buyers of fuel must be assigned
as well. All selected variables must be subject to the
limitations concerning appropriate production volume
and composition of the bioethanol production line. The
limitations are mostly veriable at the very end of the
selection process. Due to complexity in verifying the
relevant tasks, the raised problem can be considered as
NP-hard. In fact, a complex scheduling or supply chain
problem is often considered as NP-hard and dealt with the
use of modern heuristic optimisation methods [31].
A problem H is NP-hard (non-deterministic
polynomial-time hard) when every problem L in NP can
be reduced in polynomial time to H. If a solution for H
takes one unit time, it can be used to solve L in polynomial
time.

1207
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Swarm-Assisted Investment Planning...
The presented optimisation problem can be solved
twofold. It can be modelled as a routing problem in a
polytree, i.e., a directed decision tree (Fig. 3). Nodes of
this graph represent particular devices and locations of the
bioethanol plant. Edges of the graph represent total cost
transitions. The solution in this case would be a path from
an initial node to nal node, where all subsequent nodes
on the path represent particular pieces of equipment, plant
locations, and buyer/seller locations to be chosen.
The second possibility is to conduct a search for the
optimum point in a multidimensional space, based on an
objective function. Each of the variables determine the
choice of a particular device for the appropriate stage of
production, and choice of locations on the map for beet
suppliers and bioethanol buyers, which adds up to 37
optimisation variables to be determined. Both approaches
have been tested to determine the most suitable.
(10)
… where eqn indicates a chosen piece of equipment for a
particular production stage, locSB indicates the location of
a raw material seller from the map, and locBE indicates the
location of a bioethanol buyer on the map. All variables
in this case are integer numbers indicating numbers of
particular items from the list or the map.
Ant Colony Optimization
ACO [32] is one of the rst optimisation techniques
inspired by the intelligence of animal swarms. The
source of inspiration was the behaviour of ant colonies,
especially the mechanism of their communication. The
ACO algorithm is mainly used in graph problems where it
has proven to be highly efcient.
The ant system algorithm is one of the basic algorithms
derived from the ACO techniques group. The algorithm by
which the optimisation process is carried out is as follows:
1. Assign m agents to the start node.
2. Assign a certain initial amount of pheromone to each
edge of the graph.
3. Construct a path for each agent.
4. Update the amount of pheromone on each edge.
5. If T iterations were performed, end the processing;
otherwise go to Step 3.
6. Save the best path.
As a well-known algorithm, ACO has found many uses
in complicated optimisation tasks. There are numerous
examples of job sequencing and operation machine
allocation done by the ACO algorithm as well [33].
River Formation Dynamics
One of the latest methods in the eld of swarm
intelligence is the river formation dynamics algorithm
[34]. The principle of its operation is to imitate the process
of riverbed formation. A set of drops, placed at a starting
point, is subjected to gravitational force that attracts
them to the centre of the earth. As a result, these drops
are distributed throughout their environment, seeking the
lowest point – the sea.
RFD utilizes this idea into graph theory problems,
creating a set of agents that move on the edges between
nodes, and explore the environment for the best solution.
This is accomplished by the mechanisms of erosion
and soil sedimentation. The amount of soil relates to an
altitude assigned to each node. Transition from one node to
another is carried out according to the decreasing gradient
between the nodes. This provides many benets for the
optimisation process (e.g., avoidance of local cycles). In
this sense, RFD is a gradient-oriented variant of the ACO
algorithm.
The RFD algorithm is as follows:
1. Initialise nodes
2. Initialise drops
3. Move drops
4. Erode paths
5. Deposit sediment
6. Analyse paths
7. If an end condition has been met, end processing;
otherwise go to Step 3
Drops move individually until they reach a goal or
evaporate – not being able to make a move. The probability
that drop k located in node i selects another node j is the
following:
(11)
… where:
(12)
Nk(i) represents a set of neighbouring nodes connected
to edges of the node in which drop k is located; altitude(i)
indicates the amount of soil in node i, and distance(i,j)
represents edge length between nodes i and j.
All nodes are eroded according to Equation (13). The
erosion is inversely proportional to the total length of the
route of a drop pathLengthk.
(13)
The nal step is to deposit a small amount of
sediment to all nodes in order to avoid approaching a
zero altitude, which would adversely affect the operation
of the algorithm. This amount decreases with successive
iterations of the algorithm.
Particle Swarm Optimization
PSO [35] is another example of a stochastic optimisation
method. Since its development in 1995 it has gained wide
popularity among researchers due to its robustness and
effectiveness in solving various optimisation tasks.

1208
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
In PSO a number of simple agent-particles is placed in
a search space, and each evaluates the objective function
at its current location. Each particle i in the swarm is
composed of three vectors indicating its current location
, previous best location , and velocity . The location of
the particle is indicated by a set of coordinates being the
problem solution. The PSO algorithm is as follows:
1. Randomly generate initial swarm.
2. Evaluate the tness of the next particle.
3. Compare particle tness value with its best; if the
current value is better than the best one, then set its
best to the current value and set the best location to
the current location .
4. Identify the best particle in the neighbourhood and
assign its index to the variable g.
5. Update the velocity and position of the particle.
6. If all particles have been processed, continue;
otherwise go to Step 2.
7. If an end condition has been met, end processing;
otherwise go to Step 2.
The PSO algorithm also nds many uses in production
optimisation, e.g., in biogas production [36].
Cuckoo Search
One of the newest swarm algorithms is the cuckoo
search, developed by Yang and Deb [37]. It was inspired
by the brood parasitism phenomenon seen in some species
of cuckoos, which is manifested by placing their eggs in
nests of birds of other species. The algorithm applies the
mechanism of Lévy ights to select subsequent nests,
allowing for proper balance between exploration and
exploitation of a search space. A Lévy ight is a random
walk in which the steps are dened in terms of the step-
lengths, which have a certain heavy-tailed probability
distribution, with the directions of the steps being isotropic
and random.
Each cuckoo lays one or more eggs (in a randomly
chosen nest), each representing coordinates of a point in
the search space, being the problem solution. The number
of nests is xed, and at the end of each iteration a part
of them is rejected with some probability with only the
best nests (with the best tness value) moved to the
next iteration. Those assumptions are represented by the
following algorithm:
1. Randomly generate initial population of n nests xi.
2. Get a cuckoo randomly by Lévy ights and evaluate
its tness.
3. Randomly choose a nest j.
4. If tness of the chosen cuckoo is better than nest j,
replace j by the new solution.
Fig. 2. Map of available locations for the plant. Grey colour
marks suppliers of sugar beet and black – petrol stations.
Fig. 3. Simplied decision tree for the optimisation process.
Algorithm iterations 100
Average annual operating time 250 days
Average motor load of a device 5.5
Price for 1 kWh of energy 0.1 USD
Table 1. Parametres used in simulation.

1209
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Swarm-Assisted Investment Planning...
5. Abandon a fraction of the worst nests and create new
ones.
6. Keep the best solutions.
7. If an end condition has been met, end processing;
otherwise go to Step 3.
The CS algorithm has proven to be very efcient in
solving various engineering optimisation tasks [38].
Furthermore, a modied version of the CS algorithm was
previously used in selecting energy system parameters for
buildings [39].
Results and Discussion
The following tests were created to test the capabilities
of swarm algorithms in a given task. An optimisation
model has been prepared for each of the four swarm
algorithms: ant colony optimization, river formation
dynamics, particle swarm optimization, and cuckoo
search. As mentioned earlier, the problem can be divided
into two types based on the used algorithm. Therefore, the
ACO and RFD algorithms were used to solve a decision
tree problem, and the PSO and CS algorithms conducted a
multidimensional space search.
The tests were conducted on a PC with an Intel i7
procesor. Matlab software was chosen as the testing
environment. All algorithms were set to work for exactly
100 iterations.
A list of equipment (Appendix a)) characterised by
price, capacity, power, and energy efciency was created
for all of the production stages, i.e., washing, cutting,
extracting, liming, two stages of carbonation, ve stages
of evaporation, two stages of mixing, fermentation,
centrifuging, and two stages of distillation. Moreover,
a list of possible sugar beet providers and petrol station
locations (Appendix b)) with relevant characteristics
Table 2. Optimisation results of the considered algorithms for 100000 l/a bioethanol production.
Table 3. Optimisation results of the considered algorithms for 500000 l/a bioethanol production.
Table 4. Optimisation results of the considered algorithms for 1000000 l/a bioethanol production.
CS RFD PSO ACO
Equipment price [USD] 101 005 101 005 101 045 101 005
Energy costs [USD] 175 375 175 375 175 375 175 375
Transportation costs [USD] 22 000 12 240 21 040 16 160
Sugar beet costs [USD] 688 843.80 688 843.80 747 887.50 688 843.80
Revenue [USD] 61 421.58 61 421.58 53 004.24 61 421.58
Total costs [USD] 925 802.22 916 042.22 992 343.26 919 962.22
CS RFD PSO ACO
Equipment price [USD] 133810.00 125505.00 126650.00 136105.00
Energy costs [USD] 18040.40 175798.60 184699.80 180404.40
Transportation costs [USD] 23440.00 42080.00 60720.00 31865.00
Sugar beet costs [USD] 3740252.00 3442460.00 3442460.00 3639172.00
Revenue [USD] 320835.90 320835.90 307085.80 320835.90
Total costs [USD] 3594706.50 3465007.70 3507444.00 3666710.50
CS RFD PSO ACO
Equipment price [USD] 217290.00 191491.00 241845.00 191491.00
Energy costs [USD] 215451.20 209647.60 209647.60 214066.80
Transportation costs [USD] 198680.00 82640.00 44920.00 43480.00
Sugar beet costs [USD] 7277724.00 6884334.00 6884334.00 6884334.00
Revenue [USD] 641671.80 641671.80 641671.80 530004.20
Total costs [USD] 7267473.40 6726440.80 6739074.80 6803367.60

1210
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
was created. All other factors used in the simulations are
presented in Table 1.
The values chosen are not representative of a real
bioethanol plant. The model presented is more a proof of
successful use of an optimisation system for the said task
rather than an example of a proper plant model. In this
case, complexity and type of data are more important than
the values themselves.
Several tests have been conducted for different
expected annual bioethanol production – 100,000 l/a,
500,000 l/a, and 1000,000 l/a – for which the results are
presented below in Tables 2, 3, and 4, respectively.
As shown in Table 2, the lowest costs were achieved
by the RFD algorithm, whereas the highest were obtained
by the PSO algorithm. However, overall differences
between particular solutions is very little, and equals in
the extreme case $76,301.04 US. The small difference
between all given total costs components is a result of the
limited possibility to choose from sugar beet sellers and
bioethanol buyers, as well as a clear idea of the amount
of equipment to buy. The small annual bioethanol demand
inuences the necessity for low device capacity and sparse
transportation.
Tables 3 and 4 present more diverse results, as the
annual bioethanol demand is much higher, which in turn
opens more possibilities. In these cases the RFD algorithm
achieved the best results as well. There is a much higher
difference between equipment costs, because the number
of required devices increased with capacity to meet the
annual bioethanol demand. In all cases the energy costs
are very similar because there is little difference between
particular device efciencies.
The most notable dissimilarity occurs in the case
of sugar beet costs. With increased annual demand,
the required sugar beet supply increases as well, which
makes the choice of particular beet supplier and plant
location susceptible to high changes in costs. As heuristic
optimisation methods, the presented algorithms are not
bound to present only the optimal solutions, therefore
a small change on the location map may result in a big
change in costs of buying and transporting sugar beets.
Agents of the space-searching algorithms, such as the PSO
and CS, can in some cases make too large a step and omit
the best solution. As is apparent, this is the case, since
those two algorithms presented the worst results in most
instances.
The RFD and ACO algorithms achieved the best overall
results, with the RFD algorithm being considerably better.
This is due to the fact that the RFD algorithm applies the
gradient-oriented method of selecting subsequent nodes.
It helps to avoid local cycles, and reinforces the best paths
while still possessing the possibility of nding different
routes. In the case of the other two algorithms, the CS
algorithm surpassed the PSO in accordance with numerous
other studies that compared those two algorithms.
This proves that the RFD and ACO algorithms
are characterised in this case by the best convergence
and robustness. To better illustrate this, a convergence
comparison has been made. Fig. 4 presents how the four
tested algorithms converged throughout the optimisation
process. It can be stated that the RFD algorithm achieved
the best result in this case. However, it needed more
iterations to reach its optimum. All other algorithms
converge at a considerably higher rate.
A nal conclusion can be drawn from the presented
result. That is, the two algorithms that solved the problem
as a graph (ACO and RFD) obtained better results than
the other two. It follows from this fact that the investment
planning problem that was formulated should be
considered as a directed graph problem. By this means, the
best results can be achieved. Moreover, because of their
robustness, swarm algorithms t into the given problem
very well. They can easily adapt to changes in the model,
making the investment planning tool very versatile.
Conclusions
This article presents a comparison of four swarm
algorithms – ant colony optimization, river formation
dynamics, particle swarm optimization, and cuckoo
search – as they relate to the task of rst-generation
bioethanol plant investment planning. The formulated
planning process consists of equipment and plant location
choice. With the assistance of those algorithms, additional
optimisation for the planning process could be performed.
Two approaches for the given problem have been
presented. One of them focuses on solving a polytree
problem by sequentially selecting all needed pieces of
equipment and the plant location according to cost criteria.
The other is formulated as a multidimensional space
search, where a sought point indicates indexes of given
equipment, their number, and plant, sugar beet seller, and
bioethanol buyer locations. The rst approach was solved
with the ant colony optimization and river formation
dynamics algorithms, and the later with the particle swarm
optimization and cuckoo search algorithms.
To solve the given problem, a list of equipment with its
characteristics (price, power, efciency, and capacity) was
Fig. 4. Normalised tness function convergence throughout the
optimisation process for all considered algorithms.

1211
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Swarm-Assisted Investment Planning...
Appendix a). Parameters of matched bioethanol plant equipment for each production stage.
Stage Price
[USD]
Capa-
city
Electrical
efciency
[%]
Power
[kW]
Process
efciency
[%]
Washing
4500
12 t/h
78
2.2 88.2
4515 80
4670 81
4545 82
4590 82
Cutting
2400
2 t/h
74
1.5
~100.00
3500 82.8
3545 79
3615 79
3570 79
5673 3 t/h 76 1.5
5783 75
4500
3 t/h
86
44650 85.5
4560 86
Extrac-
tion
1735 900 l
- - 113.73
1814 1100 l
1848 1250 l
2024 1550 l
2118 1950 l
2387 2550 l
2417 3750 l
3929 5650 l
5396 7550 l
5920 9450 l
6528 11350 l
6969 13200 l
9609 15100 l
11905 18900 l
14225 22700 l
17654 37850 l
20280 45400 l
23327 56750 l
27131 75700 l
Liming
1735 900 l
-- 106.23
1814 1100 l
1848 1250 l
2024 1550 l
2118 1950 l
2387 2550 l
2417 3750 l
3929 5650 l
5396 7550 l
5920 9450 l
6528 11350 l
6969 13200 l
9609 15100 l
11905 18900 l
14225 22700 l
17654 37850 l
20280 45400 l
23327 56750 l
27131 75700 l
Carbona-
tion
11500 5500 l/h
- - 92.19
13800 6000 l/h
16000 7000 l/h
18500 8000 l/h
20000 9500 l/h
24000 9800 l/h
Evapora-
tion
10000 5000 l
- - 78.25
35000 7000 l
45000 8500 l
65000 10000 l
80000 20000 l
100000 50000 l
Mixing
1700
580 l
79
1.5
~100.00
1710 79
1740 74
2500
2200 l
79
3
2375 85.5
2535 82.5
2400 83
2410 84
2350 84.6
3500
2800 l
83
5.5
3300 87.9
3330 86
3510 85.5

1212
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
prepared, along with a map of possible bioethanol plant
locations, sugar beet suppliers, and petrol stations.
As a result of the tests, we can conclude that the best-
performing algorithm is the river formation dynamics
algorithm. Total costs of the investment resulting from
its optimisation were slightly lower than those obtained
from the ant colony optimization algorithm. Generally, all
presented swarm algorithms produced satisfying results
and are suitable to solve the bioethanol plant design
planning problem.
The polytree approach to the given problem allowed
us to obtain better results. The ACO and RFD algorithms
have chosen less expensive and more efcient sets of
equipment, and more protable plant locations. This
indicates that the given problem should be approached in
this way in the future.
Swarm algorithms are characterised by their robustness,
and can easily adapt to much more complicated models,
which makes the presented tool very versatile. This
suggests that adding more detail and subjecting the model
to more objective functions and constraints is possible,
and can produce satisfying results. Just by the addition of
another variable that represents a different characteristic,
and its inclusion in an objective function, the model can be
easily expanded. The main drawback of the presented test
is that the model is based on randomized characteristics
of a plant. With more accurate data such a system could
be compared with actual bioethanol plants to determine
to what extent the optimization can benet real plant
Appendix b). Information about sugar beet sellers and bioethanol
buyers used in the location map in Fig. 2.
Type Position
on the map
Price
[USD]
Max
amount
Min
amount
Beet seller
B10 586.25/t 12 000 t 2 400 t
C1 586.25 13 000 2 600
D19 670 16 000 3 200
I2 670 18 000 3 600
K10 636.5 13 000 2 600
O13 619.75 11 000 2 200
R20 636.5 14 000 2 800
S8 619.75 12 000 2 400
Bioethanol
buyer
A5 6.365/l 600000 l 120 000 l
B15 6.7 800 000 160 000
E1 6.365 700 000 140 000
G6 7.705 1 600 000 320 000
G11 7.035 1 000 000 200 000
I19 7.035 1 100 000 220 000
N7 7.37 1 200 000 240 000
P16 6.7 900 000 180 000
Q1 7.37 1 300 000 260 000
T2 7.705 2 000 000 400 000
Fermen-
tation
1735 900 l
- - 148.57
1814 1100 l
1848 1250 l
2024 1550 l
2118 1950 l
2387 2550 l
2417 3750 l
3929 5650 l
5396 7550 l
5920 9450 l
6528 11350 l
6969 13200 l
9609 15100 l
11905 18900 l
14225 22700 l
17654 37850 l
20280 45400 l
23327 56750 l
Centri-
fugation
2500
3 t/h
82.8
1.5
65.08
2545 79
2600 79
2575 74
5000
12 t/h
85.5
4
5180 82.5
5030 83
5070 84
5040 84.6
10000
20 t/h
87.9
7.510145 86
10350 85.5
20000 100 t/h 89.4 11
20100 89.8
30000 200 t/h 90.6 15
Distilla-
tion
5000 640 l
- - 36.5
7500 1220 l
9500 1450 l
9800 1800 l
10500 20000 l
11500 2300 l
13500 2500 l
16000 3200 l
Appendix a). Continued.

1213
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Swarm-Assisted Investment Planning...
developers. Extension to this problem should be a topic
for future research.
Nomenclature
Pi Power of i-th device [kW]
RBE Revenues from the sale of bioethanol [USD]
CSB Costs associated with transportation of beet
[USD]
CBE Costs associated with transportation of bioethanol
[USD]
Purchase price of beet from i-th seller [USD·t−1]
Amount of beet purchased from i-th seller [t]
Selling price of bioethanol for i-th recipient
[USD·l−1]
Amount of bioethanol sold to i-th recipient [l]
CE Costs associated with the purchase of executive
devices [USD]
ci Price of i-th device [USD]
ki Number of i-th devices [–]
Effciency of i-th device [%]
Distance to i-th seller [km]
Distance to i-th recipient [km]
tW Average annual operating time [d/y]
ni Average motor load of i-th device [–]
Qel Annual electricity consumption [kWh·y−1]
G Price for 1 kWh of energy [USD]
CEN Costs associated with the purchase of electricity
[USD]
C Total costs [USD]
CSBp Costs associated with the purchase of sugar beet
[USD]
References
1. BAENA L., GOMEZ M., CALDERON J. Aggressiveness of
a 20autoparts: I behavior of metallic materials and evaluation
of their electrochemical properties. Fuel 95, 320, 2012.
2. PARK S.H., KIM H.J., SUH H.K., LEE C.S. Atomization
and spray characteristics of bioethanol and bioethanol
blended gasoline fuel injected through a direct injection
gasoline injector. International Journal of Heat and Fluid
Flow 30, 1183, 2009.
3. NAIK S., GOUD V.V., ROUT P.K., DALAI A.K. Production
of rst and second generation biofuels: A comprehensive
review. Renewable and Sustainable Energy Reviews 14,
578, 2010.
4. SALAZAR-ORDONEZ M., PEREZ-HERNANDEZ
P.P., MARTIN-LOZANO J.M. Sugar beet for bioethanol
production: An approach based on environmental
agricultural outputs. Energy Policy 55. Special section: Long
Run Transitions to Sustainable Economic Structures in the
European Union and Beyond, 662, 2013.
5. GOH C.S., LEE K.T. A visionary and conceptual
macroalgae-based third-generation bioethanol (TGB)
biorenery in Sabah, Malaysia as an underlay for renewable
and sustainable development. Renewable and Sustainable
Energy Reviews 14, 842, 2010.
6. European Commission. Proposal for a Regulation of the
European Parliament and of the Council on Support for
Rural Development by the European Agricultural Fund for
Rural Development (EAFRD). COM 627, Final/2, 2011.
7. The European parliament and the council of the European
Union. Directive 2009/28/EC of the European parliament
and of the council of 23 April 2009 on the promotion of the
use of energy from renewable sources and amending and
subsequently repealing directives 2001/77/EC and 2003/30/
EC. Official journal of the European Union 52, 16, 2009.
8. The European parliament and the council of the European
Union. Directive 2009/29/EC of the European parliament
and of the council of 23 April 2009 amending directive
2003/87/EC so as to improve and extend the greenhouse
gas emission allowance trading scheme of the community.
Official journal of the European Union 52, 63, 2009.
9. The European parliament and the council of the European
Union. Directive 2009/31/EC of the European parliament
and of the council of 23 April 2009 on the geological storage
of carbon dioxide and amending council directive 85/337/
EEC. European parliament and council directives 2000/60/
EC, 2001/80/EC, 2004/35/EC, 2006/12/EC, 2008/1/EC
and regulation (EC) No 1013/2006. Official journal of the
European Union 52, 114-135, 2009.
10. Krajnc, D. & Glavic, P. Assessment of different strategies
for the co-production of bioethanol and beet sugar. Chemical
Engineering Research and Design 87. Special Issue on
Biorenery Integration Biorenery Integration SI, 1217,
2009.
11. REDLARSKI G., PIECHOCKI J., DąBKOWSKI M.
Reducing Air Pollutant Emissions from the Residential
Sector by Switching to Alternative Energy Sources in Single-
Family Homes. Polish Journal of Environmental Studies 22,
197, 2013.
12. VUCUROVIC D.G., DODIC S.N., POPOV S.D., DODIC
J.M., GRAHOVAC J.A. Process model and economic
analysis of ethanol production from sugar beet raw juice
as part of the cleaner production concept. Bioresource
Technology 104, 367, 2012.
13. FAKHARUDIN A.S., SULAIMAN M.N., SALIHON J.,
ZAINOL N. Implementing Articial Neural Networks and
Genetic Algorithms to Solve Modeling and Optimisation of
Biogas Production in Proceedings of the 4th International
Conference on Computing and Informatics, ICOCI 2013,
121, 2013.
14. WOLF C., MCLOONE S., BONGARDS M. Biogas plant
optimization using genetic algorithms and particle swarm
optimization in IET Irish Signals and Systems Conference
(ISSC 2008) (Galway, Ireland), 244, 2008. DOI:10.1049/
cp:20080670.
15. WEI X., KUSIAK A. Optimization of Biogas Production
Process in a Wastewater Treatment Plant in Proceedings
of the 2012 Industrial and Systems Engineering Research
Conference (Orlando, FL), 1, 2012.
16. KOCOLOSKI M., GRIFFIN W.M., MATTHEWS H.S.
Impacts of facility size and location decisions on ethanol
production cost. Energy Policy 39, 47, 2011.
17. CELLI G., GHIANI E., LODDO M., PILO F., PANI S.
Optimal location of biogas and biomass generation plants
in Universities Power Engineering Conference, 2008.
UPEC 2008. 43rd International, 1, 2008. DOI:10.1109/
UPEC.2008.4651490.
18. LEDUC S., STARFELT F., DOTZAUER E., KINDERMANN
G., MCCALLUM I., OBERSTEINER M., LUNDGREN J.
Optimal location of lignocellulosic ethanol reneries with
polygeneration in Sweden. Energy 35, 2709, 2010.

1214
Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy • Author Copy
Redlarski G., et al.
19. MAS M.D., GIAROLA S., ZAMBONI A., BEZZO F. In
20th European Symposium on Computer Aided Process
Engineering (Eds Pierucci S., Ferraris G.B.) 97, Elsevier,
2010. DOI:10.1016/S1570-7946(10)28017-3.
20. ZHANG J., OSMANI A., AWUDU I., GONELA V. An
integrated optimization model for switchgrass-based
bioethanol supply chain. Applied Energy 102. Special Issue
on Advances in sustainable biofuel production and use –
XIX International Symposium on Alcohol Fuels – ISAF,
1205, 2013.
21. OSMANI A., ZHANG J. Economic and environmental
optimization of a large scale sustainable dual feedstock
lignocellulosic-based bioethanol supply chain in a stochastic
environment. Applied Energy 114, 572, 2014.
22. BAI Y., HWANG T., KANG S., OUYANG Y. Biofuel
renery location and supply chain planning under trafc
congestion. Transportation Research Part B: Methodological
45, 162. ISSN: 0191-2615, 2011.
23. AKGUL O., SHAH N., PAPAGEORGIOU L.G. Economic
optimisation of a UK advanced biofuel supply chain.
Biomass and Bioenergy 41, 57. ISSN: 0961-9534, 2012.
24. BALAMAN S.Y., SELIM H. A decision model for cost
efective design of biomass based green energy supply chains.
Bioresource Technology 191, 97. ISSN: 0960-8524, 2015.
25. SAMER M. A software program for planning and
designing biogas plants. Transactions of the ASABE 53,
1277, 2010.
26. KARUPPIAH R., PESCHEL A., GROSSMANN I.E.,
MARTIN M., MARTINSON W., ZULLO L. Energy
optimization for the design of corn-based ethanol plants.
AIChE Journal 54, 1499, 2008.
27. KOSTIN A., GUILLEN-GOSALBEZ G., MELE F.,
BAGAJEWICZ M., JIMENEZ L. Design and planning of
infrastructures for bioethanol and sugar production under
demand uncertainty. Chemical Engineering Research and
Design 90, 359, 2012.
28. TONG K., GONG J., YUE D., YOU F. Stochastic
Programming Approach to Optimal Design and Operations
of Integrated Hydrocarbon Biofuel and Petroleum Supply
Chains. ACS Sustainable Chemistry & Engineering 2, 49,
2014.
29. CASTILLO-VILLAR K.K. Metaheuristic Algorithms
Applied to Bioenergy Supply Chain Problems: Theory,
Review, Challenges, and Future. Energies 7, 57. ISSN:
7640-7672, 2014.
30. SANTEK B., GWEHENBERGER G., SANTEK M.I.,
NARODOSLAWSKY M., HORVAT P. Evaluation of
energy demand and the sustainability of different bioethanol
production processes from sugar beet. Resources,
Conservation and Recycling 54, 872, 2010.
31. YANG X.-S., DEB S., FONG S. In Networked Digital
Technologies (Ed Fong, S.) 53–66 (Springer Berlin
Heidelberg). ISBN: 978-3-642-22184-2. DOI:10.1007/978-
3-642-22185-9_6, 2011.
32. DORIGO M., BIRATTARI M., STUTZLE T. Ant colony
optimization. Computational Intelligence Magazine, IEEE
1, 28, 2006.
33. PRAKASH A., TIWARI M.K., SHANKAR R. Optimal job
sequence determination and operation machine allocation in
exible manufacturing systems: an approach using adaptive
hierarchical ant colony algorithm. Journal of Intelligent
Manufacturing 19, 161. ISSN: 1572-8145, 2008.
34. RABANAL P., RODRIGUEZ I., RUBIO F. In Unconventional
Computation (eds Akl S., Calude C., Dinneen M., Rozenberg
G. & Wareham, H.) 163–177 (Springer Berlin Heidelberg).
DOI:10.1007/978-3-540-73554-0_16, 2007.
35. POLI R., KENNEDY J., BLACKWELL T. Particle swarm
optimization. Swarm Intelligence 1, 33, 2007.
36. WEI X., KUSIAK A. Optimization of Biogas Production
Process in a Wastewater Treatment Plant in Industrial and
Systems Engineering Research Conference, 1, 2012.
37. YANG X.-S., DEB S. Cuckoo Search via Lévy ights in
Nature Biologically Inspired Computing, 2009. NaBIC
2009. World Congress on (2009), 210–214. DOI:10.1109/
NABIC.2009.5393690, 2009.
38. YANG X.-S., DEB S. Cuckoo search: recent advances and
applications. Neural Computing and Applications 24, 169,
2014.
39. PIECHOCKI J., AMBROZIAK D., PALKOWSKI A.,
REDLARSKI G. Use of Modied Cuckoo Search algorithm
in the design process of integrated power systems for modern
and energy self-suffcient farms. Applied Energy 114, 901,
2014.