+ Models

Energy and Buildings xxx (2006) xxx–xxx

www.elsevier.com/locate/enbuild

Optimization of Mediterranean building design using genetic algorithms

Essia Znouda *, Nadia Ghrab-Morcos, Atidel Hadj-Alouane
Ecole Nationale d’Ingénieurs de Tunis, B.P. 37,1002 Tunis-Belvédère, Tunisia
Received 5 April 2005; received in revised form 17 October 2005; accepted 20 November 2005

Abstract
While it is possible to check the energy performance of a given building by means of several available methods, the inverse problem of
determining the optimum configuration given a desired performance is more difficult to solve. In the Mediterranean region this problem is more
complex due to the following two reasons: the air-conditioning load is as important as the heating load, and the energy needs depend on a high
number of architectural parameters which have different, even contradictory, effects on summer and winter loads. In this paper we present an
optimization algorithm that couples pseudo-random optimization techniques, the genetic algorithms (GA), with a simplified tool for building
thermal evaluation (CHEOPS) for the purpose of minimizing the energy consumption of Mediterranean buildings. Since increasing the energy
performance usually requires the use of special devices resulting in a high construction cost, we also propose to use GA for the purpose of
economical optimization.
# 2006 Elsevier B.V. All rights reserved.

Keywords: Energy optimization; Cooling and heating load; Mediterranean building; Genetic algorithms

1. Introduction generally lower than the required indoor temperature, un-

insulated but high thermal capacity walls allow for the
The energy performance of a building depends on a high evacuation of the heat stored in the building during the day,
number of parameters. It is determined by its response as a leading to the reduction of air-conditioning need. One
complete system to the outdoor environment and the indoor important question is raised: what is the wall composition
conditions. Improved levels of performance require the that leads to the lowest energy consumption in both seasons?
coherent application of measures which altogether optimize The answer is not straightforward.
the performance of the complete building system. Given the In order to optimize the energy performance of the building,
number of individual attributes that have to be combined to D.A. Coley and S. Schukat proposed in [1] to use genetic
make a single building, the number of possible designs is very algorithms (GA) coupled with a dynamic thermal model, but
large, and determining the most efficient one is a complex they considered only the need for heating energy. In the present
problem. paper, we propose to use these techniques to find the
Optimization of building energy performance is more appropriate trade-offs in the Mediterranean context, i.e.
complex in the case of Mediterranean buildings. While in some minimize the sum of heating and cooling loads, which we
cold European regions only heating energy consumption is shall try to compute in a simplified way in order to be able to
usually considered, the Mediterranean climate makes it quickly evaluate numerous solutions.
essential to consider both heating and cooling energy uses. Since special measures are required to reduce the energy
Varying some parameters of the building over their ranges of loads of the building, we also propose to use genetic algorithms
practical values can have opposite effects on heating and in the economic optimization of the Mediterranean buildings.
cooling energy consumptions. It is evident, for example, that an We aim to minimize the building global monetary cost,
insulated building envelope helps in reducing the heating including the construction investments, purchase price and
demand. But in summer, the outdoor night temperature being maintenance costs of the heating and air-conditioning systems,
as well as the cost of the total energy consumption through the
life cycle of the building.
* Corresponding author. Tel.: +216 98 485 195. The main characteristics of the two sided problem are: a large
E-mail address: [email protected] (E. Znouda). multi-dimensional space to be searched, a range of different
0378-7788/$ – see front matter # 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.enbuild.2005.11.015
ENB-2114; No of Pages 6
+ Models

2 E. Znouda et al. / Energy and Buildings xxx (2006) xxx–xxx

2.1. Simplified evaluation tool

Nomenclature
Some anterior studies [3] have shown the increased
A surface of the building (m2)
importance of the dynamic behavior of the building envelope
C cooling load (Wh)
in the Tunisian context, and more widely in the Mediterranean
Cea energy operating cost (TND)
one. In the absence of free gains (solar and internal gains) the
Ceq,cool purchasing and installation cost of the cooling
heating need is uniquely due to the losses through the dynamic
equipments (TND)
system, noted Lh, which result from the difference between the
Ceq,heat purchasing and installation cost of the heating
indoor and the outdoor temperatures. This load is partly
equipments (TND)
satisfied by the useful gain Qu which is a fraction of the gross
Cglobal global monetary cost (TND)
free gains Qg that contributes to increasing the indoor
Cmaint preventive maintenance cost (TND)
cool temperature until the set point. The heating needs H are then
Cea cooling energy cost (TND)
heat given by:
Cea heating energy cost (TND)
DHC degree-hours corresponding to a cooling load H ¼ Lh
Qu (1)
during the hot season (8C h)
DHh degree-hours corresponding to a heating load The gross free gains Qg are the sum of the internal free gains
during the cold season (8C h) and the whole amount of solar gains received by the building
F solar fraction envelope. The fraction of useful provisions depends on the
Fe1 solar factor importance of the crude provisions in comparison with losses,
H heating load (Wh) on the discrepancy schedule between heat losses and solar
ielec increase rate of electricity price (%) provisions, as well as on building inertia.
igas increase rate of gas price (%) All the same, the air-conditioning load C is given by:
I basis investment (TND) C ¼ Lc þ Qeff (2)
Iconst construction investment (TND)
L1 length of the south facade (m) with Lc being the air-conditioning load for the dynamic system
Lc losses through the dynamic system during cool- in the absence of solar provisions and Qeff the effective gain or
ing season (Wh) part of free gains that induce air-conditioning needs [2]. This
Lh losses through the dynamic system during heat- amount is generally lower than the gross free gains.
ing season (Wh) Determining the dynamic-mode loads as well as the useful
Lref,c reference cooling load (Wh) and effective provisions necessitates the use of a tool for energy
Lref,h reference heating load (Wh) assessment. There are two classes of tools used for building
N populations size energy evaluation: the dynamic simulation tools based on
Qeff effective gains (Wh) physical principles and models, and the simplified procedures.
Qg gross free gains (Wh) The first ones provide accurate and detailed results but running
Qu useful gains (Wh) a simulation takes generally several minutes, and the
S solar overheating factor preparation of the input data several hours or days.
V volume of the building (m3) In order to find the optimal design of a building, we have to
Vi fraction of the glazed surface on the facade i (%) compare the energy performance of a large number of
configurations, which needs the computation of the heating
and cooling loads for each of them. In the optimization
approach proposed in the present study, we use the simplified
procedures that are more straightforward and easier.
variable types and a non-linear objective function. The results The chosen tool is the simplified method CHEOPS developed
presented in this paper demonstrate that using genetic algorithms by N. Ghrab-Morcos [2]. It has the important advantages of:
to solve such problems is a good alternative that allows us to
identify not only the best design, but a set of good solutions. - Being well adapted to the Mediterranean climatic conditions,
- Allowing for the calculation of both cooling and heating loads
2. The energy consumption problem using approximately the same method,
- Requiring little expertise.
In cold countries there is not a real need for summer air-
conditioning except where internal gains are high such as Using simple correlations, this tool allows for determining
concert halls or opera houses. Therefore, in [1] the fictitious the dynamic losses on the basis of those computed for
load of air-conditioning due to the possible overheating was permanent state:
introduced in the objective function in the form of penalty. Our
situation being different, in the present work, the objective Lh ¼ a  DHh (3)
function is taken as being the sum of the heating and air-
conditioning energy loads. Lc ¼ d  DHc þ e  DH2c (4)
+ Models

E. Znouda et al. / Energy and Buildings xxx (2006) xxx–xxx 3

DHh: degree-hours corresponding to a heating load during the Table 1

Set of the considered solar protection devices
cold season; DHc: degree-hours corresponding to a cooling load
during the hot season; and a, d and e: the correlation factors. Orientation South East/west North
The useful and effective gains are calculated on the basis of N8 1 2 3 4 5 1 2 1 2 3
the gross free gains due to the internal provisions and the solar Fe1 winter 0.9 1 0.9 0.8 0.7 0.8 0.9 1 0.9 0.8
radiation absorbed by the building envelope: Fe1 summer 0.8 0.8 0.7 0.6 0.5 0.6 0.7 1 0.9 0.8

Qu ¼ F  Lref;h (5)

Qeff ¼ S  Lreff;c (6) increase in the heating load. This means that the shadowed
portion of the glazed area should be as large as possible in
Lref,h: a reference heating load; Lref,c: a reference cooling load; summer and as low as possible in winter. Knowledge of the
and F and S are, respectively, the solar fraction of the heating shaded part is necessary to compute the gross solar gains.
season and the solar overheating factor for the cooling season. The development of the simplified tool CHEOPS [4]
They are related to the normalized gross free gains of the two included consideration of the effect of different shadowing
seasons by the means of correlations determined on the basis of devices. The solar factors Fe1 are defined as the ratios of the
a large number of detailed simulations. More details can be received solar radiation in the presence of the shadowing device
found in [2]. over the radiation that would be received in its absence. Their
average values over both the heating and the cooling seasons
2.2. Design variables have been calculated for different overhangs and shadowing
devices.
The losses across the envelope and the gross free gains Referring to this work, we selected five configurations that
depend on the lateral surface of the building, the type of used will be tested on the south facade, two configurations for the
partitions as well as glazed surfaces on each of the façades. The east and west façades and three others for the north one. All
shape and the dimensions of the solar protections have direct these devices correspond to an Fe1 factor with a high value in
impact on the amount of the solar free gains received by the winter and a reduced one in summer (Table 1).
glazed areas. We have defined a set of possible configurations, With such a set of parameters, the number of feasible
by combining different cases of these design variables, taken solutions reaches 13,500,000. In addition to the large
inside reasonable values. The resulting set of configurations dimension of the research domain, the considered problem is
defines the space of research of our problem. characterized by a non-linear objective function and discrete
variables of different natures. For these reasons, we shall use
2.2.1. Cell-test geometry genetic algorithms which are known to be efficient for the
While keeping a constant volume, we can vary the resolution of such combinatorial problems.
dimensions of the building envelope and its shape. As the
present work is a first approach of the optimization problem, we 3. The optimization algorithm
consider a simple cell-test having a rectangular shape with a
fixed volume V = 108 m3. The length of the south (or north) 3.1. Genetic algorithms
facade, is denoted by L1. To obtain several designs with
different dimensions, we kept the floor area constant Genetic algorithms have proved their efficiency in dealing
(A = 36 m2) and let L1 vary to take the following values: 1, with different energy optimization problems such as the
2, 3, 4, 6, 9, 12, 18 or 36 m. optimization of building thermal design and control [5–7] and
solar hot water systems [8] as well as the design of thermally
2.2.2. Composition of walls and floors comfortable buildings [1] and the control of artificial lights [9].
For the opaque partitions, we consider the most commonly These techniques belong to a class of probabilistic search
used solutions in the Tunisian dwellings. We consider two methods that strike a remarkable balance between exploration
different types of roofing (one insulated, the other not) and five and exploitation of the search space.
kinds of walls of different inertia and levels of insulation. Genetic algorithms are initiated by selecting a population of
Each of the four façades of the building is provided with a randomly generated solutions for the considered problem. They
glazed surface, whose area is a variable fraction Vi of the facade move from one generation of solutions to another by evolving
(i = 1 . . . 4 is the facade index). In order to limit the research new solutions using the objective evaluation, selection,
domain of the problem, we let this variable take one of the crossover and mutation operators. In general, genetic algo-
following discrete values: 0.08; 0.2; 0.4; 0.6 or 0.8. We also use rithms work with the solutions being represented by a code,
simple or double glazings that differ by their transmission rather than the initial variables. Typically, a solution is
coefficients and their loss coefficients. represented with a string of bits (also called chromosome).
Each bit position is called gene, and the values that each gene
2.2.3. Solar protections can take are called alleles [10].
For the Mediterranean climate, an efficient solar protection A basic genetic algorithm has three main operators that are
should allow for minimizing the cooling load without excessive carried out at every iteration:
+ Models

4 E. Znouda et al. / Energy and Buildings xxx (2006) xxx–xxx

Fig. 1. Used encoding.

- Reproduction: chromosomes or solutions of the current The traditional operator of mutation is replaced by the
generation are copied to the next one with some probability immigration procedure [11]. Instead of altering the value of one
based on the value they achieve for the objective function or several genes, we randomly regenerate a reduced number of
which is also called fitness. new individuals. This reduces the probability of converging to
- Crossover: randomly selected pairs of chromosomes are local optimum. It also should be noted that the developed
mated creating new ones that will be inserted in the next algorithm uses the strategy of the ‘‘populations without
generation. doubles’’ which consists in inserting the new created
- Mutation: it is an occasional random alteration of the allele of chromosomes in the new generation only when they are
a gene. different from those already reproduced. This strategy makes
the algorithm converge more rapidly.
While the selection operator for reproduction is useful for
creating a new generation that is globally better than the 4. Results and analysis
preceding one, crossover brings diversity to the population by
handling the genes of the created chromosomes and mutation Genetic algorithms are pseudo-random techniques. There-
introduces the necessary hazard to an efficient exploration of fore, the speed and efficiency of the developed algorithm will
the research space. It makes the algorithm likely to reach all the depend on the population size and the maximum number of
points of research space. iterations that have to be done before convergence. Several tests
Before developing a genetic algorithm, we must choose the carried out in [12] have shown that operating on a population of
encoding that will be used to represent an eventual solution of a large size (N > 200) and with a maximum number of 1000
the problem by a chromosome where the value of each variable iterations, gives good results. While varying the initial
is represented by one or several genes. The quality of the population, the algorithm converges towards all of the same
developed algorithm depends essentially on the adopted solution and the running time never exceeds few seconds. It can
encoding strategy and its adequacy to the used crossover and be considered as efficient and robust.
mutation operators, while respecting the nature of variables and The optimal solution given by the algorithm is represented
the constraints of the problem. on Fig. 2. It is of a rectangular shape with the principal facade
being 9 m of length; the south and north façades are longer than
3.2. The developed algorithm the east and west ones. It’s profitable to use insulated roofing
and walls as well as double glazing. While the glazed surface
In this work, a genetic algorithm was developed in order to has to be maximized on the south facade (V1 = 80%), the others
provide a method for obtaining a set of optimal architectural façades must be provided with small windows as far as possible
configurations. It has the basic structure of genetic algorithm (Vi = 80%). The best solar protections are those having the
developed by Goldberg [10]: choosing an encoding strategy, winter solar average factor very close to one.
initializing the algorithm by randomly generating the first This configuration corresponds to an annual energy load of
population, and then at each iteration creating from the current 7  106 kJ. It’s considered to be a good energy saving solution
population another one of a globally better quality, evaluated on since the annual energy load exceeds 20  106 kJ/year for some
the basis of the objective function value calculated here with of the configurations in the same context.
CHEOPS. The algorithm converges when the objective The good quality of this configuration is easy to justify. The
function of the best solution has the very same value during efficiency of the used design parameters can be explained by
several consecutive iterations. the means of the physical phenomena occurring on the building.
Since the problem variables are discrete and of different Nevertheless, varying some parameters value has opposite
types, we opted for real encoding in order to represent the effects on increasing the useful free gains and reducing the heat
solutions by a chromosome of twelve genes, as shown on Fig. 1. loss.
Every gene can take a value between zero and a maximum
value that depends on the range of the variable considered.
For the selection process, rather than the basic approach
called the wheel selection, we use an improved one, known as
the elitism approach. It consists of copying the best elements of
the current population and inserting them into the following
generation. It is then impossible for the best element of the new
generation to be worse than the one obtained in the preceding
iterations. The performances of the algorithm are then greatly
improved. Fig. 2. Optimal solution for energy consumption.
+ Models

E. Znouda et al. / Energy and Buildings xxx (2006) xxx–xxx 5

It is, for example, clear that having a large southern facade is

beneficial because it is the most sunny in winter and the least in
summer. But it is not interesting to have a building with a large
lateral surface because it increases the heat loss through the
envelope. The compromise is obtained with a 9 m long facade.
With nor insulated walls, the energy consumption of the same
configuration (9 m south facade with V1 = 80%, and Vi = 8% for
the other façades) is 19  106 kJ/year. This consumption is
Fig. 3. Economic optimal solution.
reduced to 18  106 kJ/year if the glazed surface on both east
and west façades is maximized. This confirms the fact that the
optimal value of a one design parameter depends on the global (ielec = 1.4% and ielec = 3.5%), four different economic
configuration of the building. scenarios were considered.
In [12] we used the developed algorithm to determine the
optimal building design in different Tunisian climatic regions. Thus, the global monetary cost to minimize is:
The result was the same except for the north–west region where
the climate is relatively cold; it was found that it is better to use Cglobal ¼ I const þ ð1 þ YÞ
buildings of compact shape. heat
 ðC eq;heat þ C eq;cool Þ þ Cea cool
þ Cea (9)

5. The economic problem The heating and cooling equipments were dimensioned on the
basis of the peak loads in steady state, i.e. those occurring for
Minimizing the energy consumption of a building requires the most severe outdoor conditions.
the use of special devices and measures (such as insulated walls
and solar protections) that bring about an extra investment cost 5.2. Results
not necessarily offset by the achieved reduction in energy
consumption. Hence, in the following, we shall be interested in In order to solve the economic problem, we use the
optimizing the economic performance of the Mediterranean developed algorithm described in Section 3.2 while considering
buildings, based on the evaluation of the global monetary cost. the global monetary cost as a selection criterion (fitness).
The obtained optimal configuration, represented on Fig. 3, is
5.1. Objective function of a more compact shape than the best configuration for the
energy problem, with insulated partitions, small glazy surfaces
We shall work on the research space considered for the and without solar protections. The number of glazing for
energy problem (see Section 2). But the objective function of windows depends on the economical scenario considered: with
the economic problem is defined as: a low rise in the gas price, it is optimal to use simple glazing; the
higher the fuel price is, the more it is profitable to use double
Min Cglobal ¼ I þ C maint þ C ea (7) glazing.
Compared to the optimal solution for energy savings, this
With: configuration ensures to carry out a monetary gain of 18% but
- I: the initial investment which includes construction costs the energy load is 32% higher.
Iconst and purchasing and installation costs of the heating The compact form corresponds to a minimal lateral surface.
and air-conditioning equipments, respectively, Ceq,heat and Hence, the losses through the building envelope are reduced.
Ceq,cool, On the other hand, the solar provisions, which are profitable in
- Cmaint: the preventive maintenance cost that is proportional to winter, are also reduced and the night refreshment, desirable in
the purchase price: summer, is limited, leading to an increase in the global energy
load. This increase is balanced out by the reduction of the
C maint ¼ Y  Ceq (8)
construction investment justifying the benefit of the compact
where Y is a multiplier factor that takes into account the form.
conversion to current value and the increase in the main- The large windows and the solar protections have proved to
tenance cost. be too expensive and their additional cost cannot be
- Cea: the energy operating cost, i.e. the expenditures due to the compensated by the increase (respectively decrease) of the
heat cool
heating Cea and cooling Cea energy consumptions during crude solar provisions profitable in winter (respectively
the life cycle of the building. summer).
These are calculated on the basis of the annual energy
loads for heating H and cooling C, taking into account the 6. Conclusion
increase in the gas price for heating and in electricity cost for
air-conditioning as well as discounting future expenses. The energy problem presented in this paper is particularly
With two values for the increase rate of gas price interesting. While it is relatively easy to find the best
(igas = 0.2% and igas = 8.2%) and of electricity cost characteristics of a building under winter or summer conditions
+ Models

6 E. Znouda et al. / Energy and Buildings xxx (2006) xxx–xxx

separately, tackling the two problems simultaneously is more optimization approach. It will also be interesting to expand the
complex. There is a trade-off that has to be done between the presented algorithm to the case of more complex and varied
two seasons requirements. buildings so as to provide the designers with a powerful new
We have developed an optimization algorithm coupling the design tool.
genetic algorithms’ techniques to the thermal assessment
simplified tool for Mediterranean buildings CHEOPS. This
algorithm is used to identify the best configurations from both References
energetic and economic points of view. It proved its
[1] D.A. Coley, S. Schukat, Low energy design: combining computer-based
effectiveness by determining the most adequate architectural optimization and human judgement, Building and Environment 37 (12)
design to the considered climate and under the fixed objectives (2002) 1241–1247.
in only a few seconds. This is due to the fact that genetic [2] N. Ghrab-Morcos, CHEOPS: a simplified tool for thermal assessment of
algorithms represent a simple and very efficient approach for Mediterranean buildings in both hot and cold seasons, Energy and
the solution of non-linear combinatorial optimization pro- Buildings 37 (6) (2005) 651–662.
[3] N. Ghrab-Morcos, Analyse et simulation du comportement thermique des
blems. Although GA find good solutions without exploring the structures architecturales vis-à-vis des apports solaires, Thèse de Doctorat,
whole space of research, yet they need the evaluation of a large Faculté des sciences de Tunis (1991).
number of building configurations. This was made possible [4] N. Ghrab-Morcos, Outil simplifié pour le calcul des coefficients regle-
with the help of CHEOPS which is a simplified evaluation tool, mentaires au Maghreb, Congrès de 1’IBPSA, Paris, Octobre 2002, pp.
160–164.
requiring little time to compute results of a good accuracy.
[5] J.A. Wright, H.A. Loosemore, R. Farmani, Optimization of building
Actually, the correlations embedded in the tool have been thermal design and control by multi-criterion genetic algorithm, Energy
developed on the basis of the results of a very large number of and Buildings 34 (9) (2002) 959–972.
cases, simulated with detailed tools, allowing the dynamic [6] T.T. Chow, G.Q. Zhang, Z. Lin, C.L. Song, Global optimization of
aspects to be represented in CHEOPS. absorption chiller system by genetic algorithm and neural network,
The algorithm presents also the big advantage of converging Energy and Buildings 34 (1) (2002) 103–109.
[7] W. Huang, H.N. Lam, Using genetic algorithms to optimize controller
not only toward the best solution but toward a set of parameters for HVAC systems, Energy and Buildings 26 (3) (1997) 277–
configurations all of a high quality and diverse enough to 282.
allow the user to choose the most adequate one to his personal [8] M. Loomans, H. Visser, Application of the genetic algorithm for optimi-
considerations that are not necessarily quantifiable. The fact zation of large solar hot water systems, Solar Energy 72 (5) (2002) 427–
439.
that the required result is a set of very good solutions (and not
[9] D.A. Coley, J.A. Crabb, An artificial intelligence approach to the predic-
the best one) means that a good evaluation accuracy is tion of natural lighting levels, Building and environment 32 (2) (1997) 81–
sufficient. 85.
This is why the use of CHEOPS is satisfactory and coherent [10] D.E. Goldberg, Genetic algorithms in search, optimization and machine
with the heuristic approach of GA, detailed evaluation methods learning, The University of Alabama, Addison-Wesley, 1989.
with very high accuracy being useless in this case. [11] A.B. Hadj-Alouane, J.C. Bean, A Genetic algorithm for the multiple-
choice integer program, Operations Research 45 (1) (1997) 92–101.
The obtained results show that the best solutions for saving [12] E. Znouda, Contribution à l’optimisation énergétique des bâtiments à
energy and saving money are quite different. It is possible to l’aide des algorithmes génétiques, Mémoire de DEA, Ecole Nationale des
combine the two problems and then carry out a multi-objective Ingénieurs de Tunis (Novembre 2003).