Artificial Intelligence and Sustainable Design — Papers from the AAAI 2011 Spring Symposium (SS-11-02)
Generation of Energy-Efficient Patio Houses: Combining GENE_ARCH
and a Marrakesh Medina Shape Grammar
Luisa Caldas
Faculty of Architecture, Technical University of Lisbon
Rua Sá Nogueira, Polo Universitário da Ajuda, 1349-055 Lisbon, Portugal
CIAUD: Centro de Investigação em Arquitectura, Urbanismo e Design
IHSIS: Institute for Human Studies and Intelligent Sciences
Email: lcaldas@fa.utl.pt
scale, addressing the traditional patio house of the Marrakech Medina. The objective is to create a new generative
system able to incorporate the essential features and spatial
topology that makes this kind of house unique, as well as
to assess and optimize the environmental behaviour of this
typology. Patio houses have been chosen for this experiment due to several reasons. First, the Medina of Marrakech and its patio houses represent an intrinsically related
structure covering the urban and the building scale, which
become in-dissociable one from the other. Second, the
complexity and diversity verified in the architectural corpus was attractive mostly because many houses configurations emerge from a systematic application of rules and
constraints that define the core of this typology. Third, the
intricate relation between urban fabric, patio houses and
the progression from the public realm to the private space
that emerges from this relation, imprints in the space fundamental ideas that may be applicable in another contexts.
The corpus selected houses are the same as the ones
used for the house grammar development (Duarte et al,
2006b). Although the data previously collected and analyzed was partially applied, new and more complex analysis stages were needed, due to specific issues related to the
consolidation of the generative process, specifically required for GENE_ARCH. From this renewed analysis
process, new relations, constraints, patterns and rules
emerged.
In previous work (Caldas, 2008) GENE_ARCH has
proved to be flexible and capable to incorporate constraints
that allow the translation of architectural concepts and intentions into a computer-based system. In this paper, the
objects of GENE_ARCH will expand from the architecture
of a singular building to the complexity of an architectural
typology. The objective is to demonstrate, by the formulation of a parameterization methodology, how to proceed to
the implementation of the formal structure and syntax of a
traditional architectural typology into GENE_ARCH. The
proposed methodology consists in five steps or stages. The
first one is the Architectural Analysis of the several subjects, from which one can characterize the typology under
study. The second stage is to establish a Initial Basic Model
that synthesises all the essential features of the architec-
Abstract
GENE_ARCH is a Generative Design System that
combines Pareto Genetic Algorithms with an advanced energy simulation engine. This work explores
its integration with a Shape Grammar, acting as GENE_ARCH’s shape generation module. The islamic
patio house typology is readdressed in a contemporary
context, by improving its energy-efficiency, and rethinking its role in the genesis of high-density urban
areas, while respecting its specific spatial organization
and cultural grounding. Field work was carried out in
Marrakesh, surveying a number of patio houses, becoming the Corpus of Design, from where a shape
grammar was generated. The computational implementation of the patio house grammar was done
within GENE_ARCH. The resulting program was
able to generate new, alternative patio houses designs
that were more energy efficient, while respecting the
traditional rules captured from the analysis of existing
houses. After the computational system was fully implemented, it was possible to realise a large number of
experiments. The first experiments kept more restrained rules, thus generating new designs that closer
resembled the existing ones. The progressive relaxation of rules and constraints allowed for a larger number of variations to emerge. Analysis of energy results
provide insight into the main patterns resulting from
the GA search processes.
1. Introduction
This paper describes research aimed at the integration of
Shape Grammars with a Generative Design System, GENE_ARCH, based on genetic algorithms and a sophisticated energy simulation algorithm (Caldas 2001, Caldas
2008). The goal was to develop a computational system for
generating novel urban and housing designs that not only
respect the formal structure of a coherent Corpus of Design, but also optimizes the environmental behaviour of
design solutions, in terms of thermal and lighting conditions.
While the urban grammar has been described elsewhere
(Duarte et al., 2006a), this paper focuses on the housing
10
sides. The bayt walls always make an orthogonal angle
with the courtyard. A direct relation can be found between
the social and economic welfare of a family, and the size
and architectural ornamentation of the patio.
The inner courtyard, with its quadrangular or slightly rectangular shape, is the common and agglutinating space of
the house, where all the social activities of the family (or
families) take place. The patio is normally surrounded by
galleries, an important element in the architecture of the
Moroccan patio houses The several rooms open to the inner patio, and it is from the patio, through the galleries, that
one can access them. Galleries not only act as a buffer
space for the bayts more private nature, but they also provide horizontal circulation to the floor above, and shade for
the openings of the rooms facing the patio.
tural corpus and from which any configuration can be derived. In the third step, all the rules and constraints to the
possible transformations applied to the Initial Basic Model
are set up. In the fourth step, the variable parameters are
defined and coded, concerning relations between rules, as
well as constraints, established in the previous step. Finally, in the fifth stage, the objective function is formulated
in order to guide the evolutionary-based optimization process.
2 Urban and housing context
Zaouiat Lakdar is one the oldest zones of the Marrakech
Medina, and its urban and architectural patterns suggest a
well defined and coherent stylistic architectural corpus,
making it an adequate subject for this research. The urban
fabric is composed mostly of patio houses, where the patio
serves as the main source of daylight and natural ventilation to the interior spaces. Contrarily to western urban configurations, where buildings get light and air from windows opened into streets, here streets are reduced to minimum width, and act mostly as physical access to houses.
The exception are the main public streets which connect
the main housing quarters with major public spaces or
buildings (i.e. Mosques, public squares). From there, there
is a ramification of the street network in semi-public
partially-covered corridors (derbs), that will lead to the
private space of the houses. This gradient between public
and private space is a constant in the structure of the urban
layout found in the Medina.
3 Patio House Shape Grammar
For the elaboration of the shape grammar of the Marrakech Medina patio house (Duarte, 2006b), a site survey
was performed of eight houses that compose the Corpus of
Design. Through the corpus analysis, it was possible to
establish a basic pattern formed by three rectangular rings
around the inner patio in the typical two floors (fig. 2).
Wing 3
Gallery
Wing 8
Gallery 8
Gallery
Wing 1
Gallery 6
Wing 6
Wing 7
Gallery
Gallery 2
Wing 2
Wing 4
Gallery 4
Y-
Gallery
Wing 5
XFig. 2: House Grammar Basic Pattern. Decomposition of the
patio houses basic structure into patio, galleries and rooms or
‘bayts’ . Right, ground floor; Left, upper floor.
Around the patio, the first ring corresponds to the galleries, the second one to the rooms (bayts), and the third
one to additional rooms. A third ring was excluded from
this study, since it is used mostly for adaptation of the rigid
patio house scheme to site specificities, and is dealt with in
what was called the Negotiation Grammar (Duarte, 2006b).
As for the two inner rings, they may have one, two, three
or four sides.
For the development of the grammar, a dimensional
analysis was undertaken in order to extract architectural
proportions - like the ratio between the length and width of
the patios - and the dimensional intervals for some important variables and parameters, in order to insure that the
grammar results are within the architectural morphology of
this typology.
Fig. 1: Aerial view of the Zaouiat Lakhdar area of the Marrakech
Medina (left), a view of a patio(center), and plan of the area under study, showing both derbs and patio houses (right).
In the patio houses of the Marrakech Medina the typology
is flexible, multifunctional and with organic features that
guide its growth and configuration. Typically with two
storeys, the patio house is composed by the placement,
around its perimeter, of long and narrow rooms called bayt.
Just like the house is the fundamental cell of the urban fabric, the bayt is the house’s elementary space. Adaptable to
all types of uses, the bayt can serve many purposes, from a
simple room to an independent family unit, in case the
proportions are enough for that purpose. By their distribution, the bayts enclosure the patio, normally in its four
11
behaviour is constrained by previously defined rules. The
variables allow the derivation of different solutions from
an initial model.
Stage 3, Rules and Constraints - Rules impose constraints to the variation of the Initial Basic Model assigned
variables. From the corpus analysis, one can explicit the
implicit rules underlying in the corpus examples common
architecture.
Stage 4: Objective Function - Finally, an objective
function is determined to guide the generative and optimization process. The proposed fitness criteria and search
mechanism of the GA selects several solutions of each
generation. The selected solutions characteristics will be
propagated to the next generations, and the evolutionary
process continues towards the optimization of the objective
function, until final solutions are reached.
Fig. 3: Houses of the corpus (left) and their Grammar formulation (right). Areas of the 3rd ring not included in this study.
One of the goals of this project was to be able incorporate the several features, complexities, and possibilities
that characterize the Marrakech Medina patio houses typology. Because Genetic Algorithms supports the evolutionary search mechanism of the GENE_ARCH, this system requires the implementation of a generative process
with specific characteristics. Caldas (2008) established the
key features and methods to the elaboration of a generative
process within GENE_ARCH. The constraints applied to
the variable parameters and the initial configuration are the
means to implement the architectural features that can stylistically characterize the several solutions generated. In
order to incorporate a building’s formal structure, a previous architectural analysis is essential, since it is from that
analysis that the basic formal structure can be revealed. In
this case, the shape grammar incorporates both the analysis
stage, and the definition of the Initial Basic Model. Taking
these issues into account, a methodology was conceived to
incorporate the traditional architectural typology under
study, using as a basis the House Grammar developed.
4.2. Corpus Analysis
The eight houses composing the Corpus are: Dar 27, Dar
33, Dar 73, Dar Dounia, Dar Frances, Dar Charifa, Dar
Hannah and Dar Foundouk. The analysis done for GENE_ARCH integration was threefold: volumetric, spatial
layout and elevation composition analysis.
4.2.1. Volumetric Analysis
The patio dimensions include the first ring (galleries). Thus
the insertion of the first ring will subtract from the width
and the length of the patio. The patio has a vital relevance
in the structure of the house, and determines the dimensions of al the other spaces. It is possible to find a proportion relation between the width (Wpatio) and the lenght
(Lpatio) of the patio. This ratio (Rpatio) is crucial for assessing the right proportions in the houses that will be generated by GENE_ARCH. Table 1 shows the patio dimensions and ratios in the Corpus.
Table 1 - Patio dimensions and ratios found in the corpus.
4 Shape Grammar link with GENE_ARCH
Paatio dimensiions and ratios
The proposed methodology to integrate GENE_ARCH
with the shape grammar is:
Stage 1, Initial Basic Model - Departing from the basic
pattern formulated for the House Grammar, a hypothesis
for an architectural structure was established. This model
followed in many aspects the House Grammar, but had to
include other aspects to allow for the emergence of new
solutions. The initial model also had to be flexible and
elastic, in terms of its geometric behaviour, to ensure a
wide variety of solutions.
Stage 2, Variables - Several features of the Initial Basic
Model formal structure are chosen to be variable. Their
Patio houses
DAR 27
12
Wpatio (m) Lpatio (m) Rpatio (Lpat./Wpat.)
5,50
7,00
1,27
DAR 33
5,50
9,40
1,70
DAR 73
3,60
4,10
1,14
DAR CHARIFA
9,03
9,03
1,00
DAR DOUNIA
10,71
8,04
0,75
DAR FRANCES
DAR HANNAH
7,80
6,87
8,00
6,01
1,03
0,88
FOUNDOUK
6,60
6,60
1,00
Sym
metry
Table 2 - Depth of the first ring (Dg), Width of the second ring
(Wr) and correspondent patio dimensions.
Rin
ng and Patiio dimensioons
Dg Wr min. Wr max.
Patio houses
Wpatio
Lpatio
(m)
-
(m)
3,00
(m)
3,50
(m)
5,50
(m)
7,00
DAR 73
1,70
1,10
3,00
2,10
3,25
2,30
5,50
3,60
9,40
4,10
DAR CHARIFA
1,90
3,00
3,00
9,03
9,03
DAR DOUNIA
DAR FRANCES
1,41
1,00
3,67
2,80
4,07
3,40
10,71
7,80
8,04
8,00
DAR HANNAH
1,00
3,00
3,93
6,87
6,01
FOUNDOUK
1,50
2,80
2,80
6,60
6,60
DAR 27
DAR 33
Symm
Sym
symmetry is not explicit (Dar 33, Dar Charifa, Dar Frances and Dar Foundouk) one can with minor changes establish the double symmetry pattern. This implicit pattern
shows that some compositional rules of symmetry or double symmetry exist in the partitioning of the second ring.
It was observed that the depth of the first ring (Dg) was, in
most cases, very similar. Table 2 shows the depth of the
galleries, considering the correspondent patio dimensions.
Symm
Fig. 4: Implicit double symmetry in Dar Charifa. Left: original
ground floor. Right: Minor changes that have to be applied to
restore full double symmetry.
Finally, the partitioning relation between the two floors
tends to be similar. This is due to structural constraints
imposed by the traditional type of construction, mostly in
clay bricks.
However, there is no direct correlation between the dimensions of the patio and the depth of the galleries, as not always the biggest patios have the deeper galleries. Regarding the second ring, a tendency was observed for the
houses with the largest patios to have the deepest rooms
(Table 2). Dar 27, the house with the smallest courtyard
and the largest wings, is the only exception. The patio elevation walls, that face the interior, are perfectly orthogonal.
Moreover, the wings of the second ring, in most of the
cases, are perfectly orthogonal. Even in cases where the
second ring has a trapezoidal shape, the angle between
walls is close to 90º. In the ground floor, a minimum of
three wings was used to enclose the patio. In the contrary,
in the upper floor the number of existing wings is more
variable.
4.2.3. Patio Elevation Analysis
The name of each patio elevation, in the context of this
analysis, is given by its correspondent wing. Figure 5
shows the nomenclature for the patio elevations of the second ring.
4.2.2. Spatial Layout Analysis
Three key types of rooms were considered: multifunctional
rooms (bayts); the entrance hall and the staircase.
The entrance room, in most cases, is located in a corner.
Dar Dounia and Dar Foundouk are the only cases where
this does not happen. The only exterior staircase can be
found in Dar 73. The staircase is always in the same wing
of the entrance hall, or in its perpendicular wing. The space
between corner rooms is generally occupied by one or two
rooms. In few cases in the corpus where the space between
corner rooms is divided in more compartments.
In terms of patterns of plan composition, a tendency for
double symmetry between the corner rooms was observed.
Figure XX shows that even in cases where this double
Fig.5: Patio elevations. Right: ground floor; Left: upper floor.
The patio elevations are fundamental in the partitioning of
the second ring because it is from them that the access of
the several rooms is made. In the eight surveyed houses,
only three (Dar Dounia, Dar Frances, and Dar Hannah) of
them have particular cases of rooms that are accessed
solely by other rooms. The elevations have in most cases
only three openings. In the 54 studied elevations, only 16
have more than three openings. Independently of the openings number the corpus shows a clear tendency for symmetry in the formal composition of each elevation. The axis of
symmetry is located in the middle of each patio elevation
13
and the elevations are composed of a central opening and
two lateral ones. The lateral openings can assume two distinct positions. The first position is right at the corner. The
second position can be located in the first quarter of the
patio length. It is the interior room layout that governs the
lateral openings position. If a corner rooms belongs to the
wing, the correspondent patio elevation will have lateral
openings positioned in the first position, and becomes a
door into that room. If not, the lateral openings will assume
the second position (Fig. 6).
From this analysis, it was possible to establish three different patio elevation patterns. Figure 8 shows the relation
between each elevation type and the correspondent wing
partitioning. Considering that the rooms always have access from the elevation openings, Elevation Type I occurs
when the corner rooms are in the same wing. Type II happens when the wing has no corner rooms and is divided in
two rooms. Finally, the Type III elevation is a derivation of
Type II but with only one room in the corresponding wing.
Type I Ele-
Type II Ele-
Type III
Fig. 8 - The three elevations types that can be derived from the
Corpus, and the corresponding wing partitioning, in plan.
4.3. Initial Basic Model
Fig.6: Positioning of the patio elevations lateral openings and its
relation with the internal layout of the second ring. Left: First
position - Dar Frances, Elevation 8. Right: Second position - Dar
Frances, Elevation 1.
The Initial Basic Model is like a ‘rubber building’ that
can be further sculpted by the GA-based search process.
Figure 10 shows the Initial Basic Model that results from a
simplification from the more complex house shape grammar. This model is a two storey house where the heights of
each floor are the same. Each floor contains four galleries,
that compose the first ring, and four wings, in the second
ring. Each wing can contain a maximum of two rooms
between the corner rooms making a maximum of 24
rooms that is contemplated in this basic model. However,
this number does not match the room maximum number
that can be generated (it will always be smaller), due to
the number of openings available in each patio elevation.
Two types of parametric elevations can be assigned to each
wing depending on the corner rooms layout composition.
Each one has three openings, that are also parametric. Figure 9 shows the parametric patio elevations. Case A is applied when the wing has two corner rooms and generates
Type I Elevations. If the wing does not have corner rooms
then the Case B will be applied. Is from this elevation that
Type II and Type III Elevations can be derived.
A relation was established between the axis of the
openings located at the second position and the patio dimensions. In average this axis is located on 0,19 of Wpatio
or Lpatio. It was also observed that the central openings
are never smaller than the lateral ones. In the limit all elevation openings have the same dimensions.
The patio elevations have a tendency to be doublesymmetric between them, that is, the elevations of parallel
wings are identical. In the houses where this symmetry is
not explicit, it is because some minor misalignment or inexistence of specific openings. Figure 7 shows two cases
one with perfect double symmetry between the several
elevations and another that is almost bi-symmetric.
Fig. 7 - Double symmetry between the patio elevations. Left:
complete double symmetry in Dar Charifa. Right: incomplete
double symmetry in Dar 33 (it lacks the yellow marked window to
become perfectly bi-symmetric).
Fig. 9: Parametric Patio Elevations. Left: Parametric Elevation
A - applied when the wing has corner rooms. Right: Parametric
14
best designs from three separate runs of GENE_ARCH.
The third group (Const) are solutions resulting from relaxing the constraints imposed by the traditional solutions and
allowing more freedom to the system, and will be shown in
figure 12.
R22R21R20R191st Floor
R18
R23
R24
R17
R13R14R15R16
R10R9 R8 R7
R6
R11
R12
R5
R1R2 R3 R4 Ground Floor
Elevation B - applied when the wing does not have corner rooms.
Fig. 10 Initial Basic Model. Left: Floor plans with room nomenclature. Center: Perspective of the model. Right: Exploded view.
Fig. 11: From left to right: Random solutions Rand 1(#5), Rand 2
(#17); Optimized solutions Opt 1 (#1), Opt 2 (#4), Opt 3 (#16).
It should be noted that the random solutions are also
generated within the rules and constraints of the medina
patio houses, as encoded by the shape grammar. They are
only random in the sense that they have not yet been optimized from an energy point of view.
4 Experiments
Initial experiments were performed using the shape grammar with all the constraints, as derived from the Corpus
analysis. The goal was to determine, within the same constraints, if it was possible to improve the energy performance of the buildings, or if the constraints were so limiting
that no significant improvement could be achieved. Variables were: patio dimensions and proportions, house dimensions, space layout and type of elevation (interrelated),
dimensions of windows and doors, existence or not of galleries (ground and first floors), and gallery depth. Patio
proportions were within the grammar limits, and so were
minimum and maximum dimensions for each element,
from general house dimensions to windows and doors. In
terms of galleries, in the ground floor all four galleries had
to exist, to serve as access to first floor rooms, but the first
floor galleries were allowed to exist or not. In case they
existed, their dimensions followed the grammar rules.
A second set of experiments involved relaxing some the
constraints imposed by the traditional solutions and allowing more freedom to the system. New constraints permitted
larger patio ratios (ratio between length and width), up to
2:1, allowing for longer and slimmer patios. The need for
double symmetry in opposing elevations was suppressed,
what is significant in environmental terms (for example,
south and north facades can have significantly different
requirements). However, the requirement for space layout
double symmetry was kept. The maximum window/door
width was increased to be as large as the wall where it
stood, minus two gallery depths, instead of having a limit
fixed by the grammar. Finally, the houses were additionally allowed to have only one storey. This last case is the
one displayed in this paper.
Three groups of GENE_ARCH generated patio-house solutions are presented in figure 11, with the north direction
facing upward. The first group, to the left of the image, are
random solutions (Rand), randomly selected from those
non-optimized solutions initially generated by the GA. The
second group, to the right of the image, consists of optimized solutions (Opt), keeping all the constraints from the
existing medina patio houses. The solutions shown are the
5 Analyis of results
Solutions were analyzed in three different ways: 1)
Visual inspection, with renders and movies; 2) Energy
analysis; 3) Analysis of shape characteristics.
5.1 - Visual Inspection
Figure 12 displays solution Opt 2 in greater detail. The
patio has a ration of 1.29, as the depth of the rooms is
4.1m. There is only one overhang in the top floor, which is
facing south. The south and north facades are similar, as
imposed by double symmetry rules. GENE_ARCH makes
the elevations with the most percentage of glazing those
facing south (shaded and north), thus orienting the building
so that the wings with three rooms are south and north too.
Due to the rules that relate the design of the elevations with
the spatial layout, the wing with three rooms must also be
the one that has three doors, and thus a larger percentage of
openings. In the wings that have only one room (east and
west, where shading is more difficult and the risk of overheating is higher), the central opening is larger, and the two
lateral ones are just slits in the wall.
In solution Const 1 (figure 13), the system takes advantage of the relaxation of constraints. It generates a 2:1
patio ratio, higher than previously allowed. Because the 1storey volume with the wider patio provides less selfshading, it created galleries to all directions. Since double
symmetry is no longer imposed for the elevations (even
though it is for the space), it creates larger openings to the
north, which only get direct sun at fewer hours of the year,
and shade appropriately the south facade, also providing it
with smaller openings.
Initial visual inspection of the solutions also showed
that generally, the main feature that immediately emerges
15
is that the optimized solutions have a much larger size, in
relation to the initial ones (which are somewhat closer in
aspect to the existing houses, as seen in figure 1), even
though the overall proportions remain similar.
man discomfort. However, it is rather complicated and
non-intuitive to compare solutions in terms of discomfort,
which has to be expressed both in hours and degrees. It
was then assumed that the offset between comfort conditions and actual conditions in the building could also be
expressed in terms of how much energy would be required
to be input into the building to make it comfortable, both in
terms of heating, cooling and lighting.
Results are shown in table 3. Final results are expressed in
Energy Use Intensity (EUE, in kWh/m2), which means
energy consumption per unit area, so that solutions can be
comparable despite their differences in area (since area was
not a constraint in these experiments). On average, optimized solutions reduced energy needs by an average of
62%, while complying to the same shape grammar rules
(figure 14). The several optimized solutions have very
similar EUE levels.
Table 3: Energy needs (in kWh) for random and optimized
solutions, following the same shape grammar rules.
Lighting
Rand 1 4396
Rand 2 3839
Opt 1
16940
Opt 2
11781
Opt 3
14009
Const 1 6741
Fig. 12: Solution Opt 2. Top: Exterior view, showing south overhang, and plan. Bottom: Larger glazing areas face south (shaded
by overhang) and north. Three-room wings also become south
and north. West elevation has only slits as the lateral openings.
Heating
Cooling Tot. Energy Area (m2)
2872
2432
11195
8909
9759
11781
21775
23709
37044
28780
33264
24120
34272
35174
77341
58846
67905
50350
159
132
533
389
467
328
kWh/m2
215
266
145
151
145
154
The reduction in energy use intensity was mostly
achieved by reducing the cooling loads in the houses. As
can be seen in figure 15, a reduction in the percentage of
cooling loads was very significant, improving from representing about 80% of the total loads for achieving comfort
conditions (cooling, heating, lighting), to only 60%. That
means the optimized designs are more robust in terms of
resisting to overheating, even if that stress is still the highest, due to climate characteristics. On the other hand, lighting loads increased from 15% in the random solutions to
25% in the Opt solutions, showing that the houses also
became a little darker, in the process of protection from
overheating. That does not happen with solution Const 1,
where the high form factor (relation surface/volume) promotes better daylight use. Heating loads also increased
form 9% in the random solutions, to 17% in the Opt solutions, and 28% in the Const 1 solutions (here the increased
for factor working to its disadvantage). It should be noted
that, in desert-like climates, night temperatures can be very
low.
Fig. 13: Solution Const 1, with constraints relaxed. Left-side:
Exterior view, showing overhang in all directions, and section
with plan. Right-side: Top - North facade. Bottom - South facade
at the opposite end of the patio, with less openings.
5.2 - Energy Analysis
Energy analysis was based on the following assumptions:
in the context of the Marrakech Medina, poor environmental design would not result on energy consumption, as
there would be no systems to consume energy, but in hu-
16
Fig. 14: Energy Use Intensity (energy per unit area) for the random and optimizes solutions. A 61% average reduction was
achieved, respecting the shape grammar.
Fig. 16: Area of the random and optimized solutions.
In terms of the optimized solutions, the patio proportion ratios remain very close to the random solutions, and
showed no impact in the energy consumption of the solution. The factor that most closely correlated with the decreased energy consumption of the optimized solutions
(Opt). In fact, all the Opt solutions were discovered to have
a form factor of 0.29, even though they have quite different
appearances (figure 17).
Table 4: House dimensions and plot size (including patio area)
Rand 1
Rand 2
Opt 1
Opt 2
Opt 3
Const 1
Fig. 15: Energy Use Intensity (energy per unit area) for the
random and optimizes solutions. A 61% average reduction
was achieved, respecting the shape grammar. Cooling is
the most significant end use in all cases.
Length
10.1
10.4
22.0
15.4
18.0
29.6
Width
9.8
9.6
18.9
16.7
18.8
18.9
Area
159
132
533
390
466
328
Plot size
100
99
415
257
338
558
Table 5: Patio dimensions and room width
Rand 1
Rand 2
Opt 1
Opt 2
Opt 3
Const 1
5.3 - Shape Analysis
In order to understand why GENE_ARCH was proposing the solutions under study, another level of analysis was
introduced, regarding the geometrical and formal characteristics of the proposed designs. The first observation was
the general increase in size (figure 16). However, that corresponded to a very similar Construction Index (relation
between the plot area and the built area), and thus to similar construction density levels. An important discovery was
that it is possible, in this type of urban fabric, and within
the strict rules of the grammar, to decrease energy consumption levels by 60%, without decreasing construction
density. It implies that high densities can be achieved with
reasonable energy performance, by resorting to a patiohouse based urban solution. However, the significant increase in overall building dimensions was a concern.
Length
5
6
14
9
11
21
Width
4
5
11
7
10
11
Area
20
33
149
63
110
230
Room width (m)
2.8
2.1
4.1
4.1
4.1
4.1
Table 6: Construction index, patio ratio and form factor
Rand 1
Rand 2
Opt 1
Opt 2
Opt 3
Const 1
17
Construction
Index
1.6
1.3
1.3
1.5
1.4
0.6
Patio Ratio
1.1
1.1
1.3
1.3
1.1
2.0
Form Factor
0.35
0.40
0.29
0.29
0.29
1.16
An important discovery was that it is possible, in this
type of urban fabric, and within the strict rules of the
grammar, to decrease energy consumption levels by 60%,
without decreasing construction density, implying that high
densities can be achieved with reasonable energy performance, by resorting to a patio-house based urban solution.
Acknowledgements
The author would like to acknowledge the contribution of
Luis Silveira dos Santos to the analysis stage of the work
presented in this paper.
References
Bouchlaghen, N. 2000. Optimising the design of building envelopes for thermal performance, Automation in Construction 10:
101– 112.
Caldas, L. 2002. Evolving three-dimensional architecture form:
an application to low-energy design, in: J. Gero (Ed.), Artificial
Intelligence in Design, Kluwer Publishers: 351–370.
Caldas, L., Norford, L., Rocha, J., 2003, An Evolutionary Model
for Sustainable Design, Management of Environmental Quality:
An International Journal, Volume 14, Number 3, pp. 383- 397
Caldas, L. 2006. GENE_ARCH: An Evolution-Based Generative
Design System for Sustainable Architecture, in Lecture Notes in
Artificial Intelligence, Lecture Notes in Computer Science Series,
I.F.C. Smith (Ed.): EG-ICE, LNAI 4200: 109-118
Caldas, L. 2008. Generation of Energy-Efficient Architecture
Solutions Applying GENE_ARCH: An Evolution-Based Generative Design System, Advanced Engineering Informatics, Vol. 22
(1): 59-70.
Couchoulas, O. 2003. Shape evolution: an algorithmic method for
conceptual architectural design combining shape grammars and
genetic algorithms, Ph.D. diss, University of Bath, UK.
Domer, B. Raphael, B., Shea, K., Smith, I.F.C. 2003 A study of
two stochastic search methods for structural control, Journal of
Computing in Civil Engineering 17 (3): 132–141.
Duarte, J., Rocha, J., Ducla-Soares, G., Caldas, L. 2006. An urban
grammar for the Medina of Marrakech, in: J. Gero (Ed.), Proceedings of the 2nd International Conference on Design Computing
and Cognition, Springer, The Netherlands, 483–502.
J. Duarte, J. Rocha, G. Ducla-Soares, A patio-house shape grammar for the Medina of Marrakech, in: Proceedings of
ECAADE’06, Volos, Greece, 2006.
Elbeltagi, E., Hegazy, T., Grierson, D. 2005. Comparison among
five evolutionary-based optimization algorithms, Advanced Engineering Informatics 19 (1) (2005) 43–53.
K. Shea, A. Sedgwick, G. Antonuntoo, Multicriteria optimization
of paneled building envelopes using ant colony optimization, in:
I.F.C. Smith (Ed.), Lecture Notes in Artificial Intelligence, Lecture Notes in Computer Science Series, EG-ICE, LNAI 4200,
Springer-Verlag, Berlin, Heidelberg, 2006, pp. 627–636.
Wang, W., Rivard, H., Zmeureanu, R. 2006. Floor shape optimization for green building design, Advanced Engineering Informatics 20 (4): 363–378.
Wetter, M., Wright, J., 2004. A comparison of deterministic and
probabilistic optimization algorithms for nonsmooth simulationbased optimisation, Building and Environment 39 (8): 989–999.
Wright, J., Loosemore, H.A., Farmani, R. 2002. Optimization of
building thermal design and control by multicriterion genetic
algorithm, Energy and Buildings 34 (9): 959–972.
Fig. 17: All optimized solutions (Opt) have Form Factor of 0.29
6. Conclusions
Results prove that it is possible for GENE_ARCH to
incorporate features of an architectural typology and incorporate them in the several solutions that emerged. In the
first set of experiments (Opt), the different configurations
obtained were diverse but stylistic coherent with the shape
grammar developed, and showed a 61% average decrease
in energy consumption, in relation to random patio houses
generated within the same shape grammar. The second set
of experiments were conducted in order to assess the impact on the performance and formal configurations, created
by the relaxation of some imposed constraints. One can
observe the relaxation applied has significant impact in the
appearance of the solutions, although it does not deteriorate the coherence of the architectural subjects. This relaxation had some minor penalty in relation to energy consumption levels.
From the analysis of the solutions generated by GENE_ARCH, it was possible to extract some patterns that
permit to understand the main factors underlaying the success of some configurations that emerge during the search
process. In particular, it was found that the Form Factor of
the patio houses showed a strong correlation with the energy consumption levels, with all the Opt solutions showing exactly the same Form Factor of 0.29, despite their
very different appearance.
18