Journal Pre-proof

Implementation and fine-tuning of the Big Bang-Big Crunch optimisation method for
use in passive building design

Florian Robic, Daniel Micallef, Simon Paul Borg, Brian Ellul

PII: S0360-1323(20)30089-5
DOI: [Link]
Reference: BAE 106731

To appear in: Building and Environment

Received Date: 13 November 2019

Revised Date: 5 February 2020
Accepted Date: 6 February 2020

Please cite this article as: Robic F, Micallef D, Borg SP, Ellul B, Implementation and fine-tuning of the
Big Bang-Big Crunch optimisation method for use in passive building design, Building and Environment,
[Link]

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2020 Elsevier Ltd. All rights reserved.

 Implementation and fine-tuning of the Big Bang-Big Crunch optimisation
method for use in passive building design

Florian Robic
Department Building Operation Maintenance and Safety, Polytech Angers, 62 avenue Notre Dame du Lac, 49000 Angers,France

Daniel Micallef, Simon Paul Borg

Department of Environmental Design, University of Malta, Msida, Malta.

Brian Ellul
Department of Mechanical Engineering, University of Malta, Msida, Malta.

Abstract

Passive building design involves consideration of a number of design variables which can be optimised with
respect to some design objective. The use of optimisation methods in building design is well established but
the implementation of the Big Bang-Big Crunch algorithm to passive building design is still as yet unexplored
in detail. The aim of this paper is to demonstrate the usefulness of the method as well as to fine-tune its
characteristic parameters for its efficient use in building design applications. Two building design scenarios
are used to have more than one test case and the number of design variables are limited to an extent that an
enumeration search can be performed in a reasonable time frame from which the optimum can be precisely
established. The Big Bang-Big Crunch algorithm is used and validated against the true solution from the
enumeration search. Results show that the α and β parameters should be set to 0 and around 0.8 respectively.
The ratio of the sample size to the number of design variables should ideally be between 2.6 and 2.9 for the
algorithm to remain efficient and at the same time successful. The paper also discusses the computational
efficiency gain of the Big Bang-Big Crunch algorithm compared to a full enumeration search. Computational
savings of more than 90% are possible.
Keywords: Passive buildings; building energy modelling; optimisation methods; Big Bang-Big Crunch
algorithm; thermal comfort

Nomenclature

α Parameter that controls the extent of the search space

β Parameter that controls the weighting for the previous best solution
DV (b) Design variable b
∆TI Model temperature range constant for a category I building as established under EN15251
∆TII Model temperature range constant for a category II building as established under EN15251
∆TIII Model temperature range constant for a category III building as established under EN15251
 Average error between BB-BC optimum and the true global optimum
i Error between BB-BC optimum and the true global optimum for a particular trial

I Fully documented templates are available in the elsarticle package on CTAN.

Preprint submitted to Journal of LATEX Templates February 12, 2020

 f Fitness value
I Number of iterations until convergence
nDV Number of discrete values of a design variable
N Population size
N (0, 1) The Standard Normal Distrubution
NDV Number of design variables
Ntrials Number of trials of the BB-BC algorithm
φ Optimisation function, minimum number of hours outside categories
φ0 True global optimum
Ψ Constant established in EN15251 for the calculation of the running mean temperature
SR Success rate
Θ Improvement in computational efficiency relative to an enumeration search
Te Average external temperature
Tmax Maximum allowable temperature for a particular category
Tmin Minimum allowable temperature for a particular category
Trm Running mean temperature
Tζ Zone temperature
x Vector of design variables

1. Introduction

Optimisation is perhaps one of the most sought after aspects in the design stage of any commercial and
industrial sector, primarily because it aims at improving on a raw concept or an initial design to improve the
5 resultant product. One such sector is the building design industry which often has to contend with various
proposals and approaches aimed at introducing passive design strategies to reduce energy consumption in
buildings. Prieto et al. [1] on passive cooling and climate responsive facades, Sun et al. on cost-effectiveness
of various passive retrofitting strategies [2], Wang et al. on passive design measures and active heating systems
[3], and Chen et al. [4] on integrating photovoltaics in architectural design of buildings are all examples of
10 research done in evaluating possible passive design strategies, with the aim of reaching the various sustainable
development targets set at EU (Capros et al. [5], Mata et al.[6]) and the wider international context. With so
many aspects to consider; building insulation (e.g. type of insulation [7], position and distribution of insulation
layers [8], and thickness of insulation [9]), building orientation (e.g. use of affordable passive solar design
on different building orientations [10] and the importance of the orientation on the overall building thermal
15 performance [11]), shading levels (Winkler et al. [12]), indoor humidity characterization (Bonello et al. [13]),
glazing (e.g. use of innovative thermochromic glazing material [14], influence of orientation and glazing size [14],
impact of window configuration on energy demand [15], and the effect of internal and external shading [16]) and
ventilation strategies (e.g. general ventilation strategies for sustainable buildings [17] or more specifically cross-
ventilation in a courtyard buildings[18]), to name a few, it often becomes difficult to select the most appropriate
20 design solution, particularity since often there are significant interactions between the parameters which may
result in conflict between different design solutions.

2
 In view of the ever increasing demands on the building and environment from both policy and clients,
modern building design has become quite dependent on computational tools such as building energy simulation.
Another challenge to the designer is the ever increasing diversity of building solutions available in the market.
25 In such instances the various design combinations might in some cases be daunting and require the simulation of
a huge number of test cases. The set of variables which might provide the best building performance in terms of
some objective such as energy performance of thermal comfort might not be trivial. This is where optimisation
techniques become handy when adopted with tools such as building energy simulation.
Optimisation techniques are always evolving with the scope of creating the ultimate optimisation method
30 where the global optimum, maximum or minimum depending on the specific case, is achieved in the shortest
time possible. As technology improves, so does the computational power which enables the utilisation of new
and complex modelling techniques. In turn, these complex models are inherently more difficult to optimise
thus the use of mathematical programming techniques is limited. Fuelled by these limitations, direct search
methodologies are opted for where perhaps the most basic and optimisation methodology is the Enumeration
35 (Exhaustive) Search (ES) where all the possible solutions in the design space are evaluated and the best one is
selected. It is one of the oldest problem solving approaches used with the help of computers ( Jürg [19]). The
ES methodology is deterministic and can be employed in a multitude of scenarios but it is time consuming and
in some cases the size of the design space prohibits the use of this methodology. In view of the Enumeration
Search shortcomings, new heuristic methodologies are proposed (see Rao [20]) which are generally based on
40 the evolution of populations following a set of predefined rules. One classic example is the Genetic Algorithm
(refer to Holland [21] and Haupt & Haupt [22]) which follows the evolution of the human genes. The candidates
forming the population are composed of strings called chromosomes and by applying a series of manipulation
techniques (operators), the chromosomes in a population evolve and eventually the optimal (best) chromosome
is identified based on a series of predefined limitations. The Genetic Algorithm technique is one of the most
45 popular population based optimisation techniques but other notable techniques exist (Rao [20]) such as the
Simulated Annealing (see Kirkpatrick et al. [23]). The latter methodology follows the annealing process of
critically heated metals where the atoms in a cooling metal form crystals with the minimum possible energy.
Particle Swarm Optimisation, incepted by Kennedy & Eberhart [24] is another iterative algorithm where the
population is a swarm of particles or a flock of birds. The swarm follows the best path found by one of the
50 particles and after an infinite number of iterations, the global optimum is reached. A similar optimisation
methodology is the Ant Colony presented in Colorny et al. [25] and Dorigo et al. [26] where the population is a
colony of ants each of which follow a path to arrive at the ultimate destination i.e. food. After each iteration,
the path is optimised and the global optimum is identified by the best path found by the ants.

1.1. The Big Bang-Big Crunch method

55 Since these modern optimisation techniques rely on the optimisation of a whole population, they are con-
sidered as computationally resource hungry although with today’s computational power, these techniques are
gaining popularity. A relatively new population-based optimisation technique, called the Big Bang-Big Crunch
(BB-BC) proposed by Erol et al. [27], is gaining popularity in various industrial sectors mainly due to its
simplicity, less computational requirements and speed in obtaining the optimum. This optimisation technique
60 mimics the popular big bang theory which tries to explain the evolution of the universe. In particular, the
BB-BC optimisations technique can be divided into two phases; the big bang where a random set of solutions

3
 are solved and the big crunch where the set of solutions is crunched into a single solution. After a set amount
of iterations, the single solution is optimised such that the global optimum is reached. Worth mentioning is
that this novel optimisation approach is used to solve various problems such as; weight reduction of space
65 trusses (see Kaveh & Talatahari [28, 29] and Camp [30]) and steel structures (Hasanebi & Azad[31]), parameter
estimation of structural systems (see Tang et al. [32]), solving optimal power flow problems with valve-point
effects (Labby & Attous [33]), optimisation of laminated composite materials (see Tabakov [34] and Azad &
Tolga [35]) , generation of the optimal inverse fuzzy control signal (see Tufan et al. [36] and automatic target
tracking (Gen & Hocaoglu [37]).

70 1.2. Optimisation techniques in passive building design

Using optimisation techniques to minimise or maximise complex design solutions is not a novelty, nor is the
fact that such numerical techniques can be used to optimise the design of buildings, such as for example, the use
of passive design solutions to improve the indoor conditions inside a building, and in fact a number of authors
addressed this issue. Chen and Yang [38] for example propose a two-stage design optimisation approach utilising
75 multiple linear regression (MLR), multivariate adaptive regression splines (MARS), and support vector machines
(SVM), examined by statistical models to investigate typical passive design strategies, such as, external wall
thermal resistance, building orientation and window W-value. Often, such optimisations follow or are carried
out using building simulation tools. Chen et al. [39] and Gou et al. [40] both use energy plus to simulate
the buildings performance in tandem with the optimisation methods carried out. Other authors have also for
80 example used building simulation in tandem with optimisation techniques for very specific solutions. Mukhtar et
al. [41] have for example studied the potential influence of building optimization and passive design strategies on
natural ventilation systems in underground buildings. Gong et al. [42] use the orthogonal method and the listing
method to evaluate the effect building seven passive design parameters, including wall thickness, roof insulation
thickness and window orientation amongst others, will have on the performance of buildings in different Chinese
85 regions. Although, the research presented earlier are good examples of optimisation processes, most building
performance optimisation related studies have utilized the Genetic Algorithm process. In its review article,
Nguyen et al. [43], explain how although the stochastic based Genetic Algorithm cannot guarantee the best
solution, it is frequently used as it can give a good solution within a reasonable amount of time. Nguyen
et al. also provide a number of reasons as to why Genetic Algorithm is so popular for use in conjunction
90 with building performance simulations, including handling of both continuous and discrete variables and the
ability to handle simultaneous evaluation of a number of individuals in a population. Research work which
used Genetic Algorithm in relation to building simulation performance include the research done by Znouda
et al. [44] on designing Mediterranean buildings given a desired performance level, the research done by Evins
et al. [45] who use a multi-objective genetic algorithm optimisation to optimise regulated carbon emissions
95 versus capital and running costs, the research done by Wang et al. [46] on applying multi-objective genetic
algorithms in green building design optimization, and work done by Wright et al. [47] on costs and occupant
discomfort, amongst others. Whilst the widely used GA optimisation methodology was successfully used in
optimising a wide variety of building energy problems [43], the relatively novel BB-BC algorithm is significantly
simpler to implement since it entails fewer and simpler operators which will be discussed in more detail in
100 section 3.4. Form a computational point of view, the BB-BC optimisation methodology is more efficient than
the GA counterpart since fewer computations are required to obtain improved generations. Nonetheless, the

4
 effectiveness and efficiency of the BB-BC algorithm as applied to building energy problems still needs an impetus
to gain more confidence in this research field. The methodology to study the effectiveness and efficiency of the
BB-BC algorithm in building energy problems is discussed in the following section. The BB-BC method as
105 proposed by Erol and Eksin [27] is an improvement over the more traditional genetic algorithm, however it has
seldomly been used in such type of work, relating to building simulation. Milajić [48] have been the first in
actually using the BB-BC method in relation to building modelling to minimize the total expenses and energy
savings due to a number of passive design features. The results presented by Milajić et al. [48] show how the
BB-BC can be used for building design. The focus of their paper is however on the results obtained via the
110 BB-BC method, rather than ensuring the correctness of the results and setting out base conditions aimed at
improving the overall algorithm when used in this context.
In this study, the novel BB-BC optimisation technique is utilised together with a building energy simulation
tool to study its potential in determining the fittest set of passive design parameters typically used in building
design to maximise indoor comfort conditions. The ideal algorithmic parameters for such an application will be
115 established. Finally some insight into the optimal combination of proposed design variables will be established.

2. Methodology

2.1. General approach

In the adopted approach, two buildings are first modeled using a building simulation tool such that different
combinations of passive design features can be tested, and assessed for their indoor thermal performance. All
120 simulations for both buildings and for all the different combinations of design variables are run in order to be
able to obtain the global minimum number of uncomfortable hours, in this case meaning that during a specific
hour the indoor temperature lies outside of the thermal comfort criteria set in EN 15251 - Indoor environmental
input parameters for design and assessment of energy performance of buildings addressing indoor air quality,
thermal environment, lighting and acoustics [49]. This is usually called an enumeration search (or brute force
125 method ) The results and performance of the BB-BC algorithm can therefore be evaluated and validated on the
basis of the enumeration search. In practice, using an optimisation method such as the BB-BC means that the
number of combinations of design variables are prohibitively too large to run.
The flowchart showing the algorithm used in this research is illustrated in Figure 1. The temperature results
for each floor Tζ are compared to the comfort ranges earlier described and the number of uncomfortable hours
130 is determined φ. Various optimisation runs with the BB-BC algorithm were carried out where the sample size
N , the search space parameter α and the centre of mass modifier β were varied. Both α and β were changed
in the range [0, 1]. For most simulations, convergence was found to be obtained after less than 7 iterations.
Convergence was noted once value of the number of hours remained unchanged. The solution for φ is then
compared to the global minimum which would have been established beforehand via the enumeration search.
135 This comparison determines whether the algorithm was successful or otherwise to determine the minimum. The
whole process is repeated for a total of 10,000 trials to ensure a high statistical significance of the percentage
success rate which will be defined in the results section.
It must also be noted that in this work only comfort associated with temperatures is used as an objective. If
multiple objectives are sought such as thermal comfort, visual comfort, costs and others are sought, the BB-BC
140 algorithm would have to be run on the basis of such objectives.

5
 Building Energy
Simulation

Temperature results
in each of the zones

Number of hours
 outside range

Selection of sample size

BB-BC algorithm with

different

Determine global
minimum of

and compare with true

quantity

is trial number =
10,000? N

Determine success rate

of BB-BC algorithm

Figure 1: Flowchart of the BB-BC algorithm applied to the optimisation of the number of comfort hours during the year.

2.2. Building Energy Simulation

In order to test and tailor the BB-BC algorithm for passive building design, the building simulation tool
EnergyPlus [50] is used within the DesignBuilder v5 platform [51]. This code has been well validated in the
past (see Adelkovic̀ et al. [52], Al-Janabi et al. [53] and Pereira et al. [54]) The two modelled buildings were
145 used to study a number of passive design parameters and their effect on thermal comfort within the modeled
buildings, specifically the number of hours the indoor environment would be considered thermally comfortable
or uncomfortable according to the European standard EN 15251 [49].
Similar to all other building energy simulation tools, EnergyPlus uses the underlying code to prescribe a set
of thermo-fluid equations to calculate the environmental conditions within a zone or a control volume under
150 investigation. The relative ease with which different design variable can be modified offers the possibility of
modifying a set amount of parameters and variables in quick succession to study the effect each design decision
has on the indoor conditions.

2.3. Problem definition

2.3.1. Building number 1

155 The first building modelled, shown in Figure 2, consists of a simple, square in shape, fully detached building
having dimensions 10m by 10m, and an internal height per floor of 3.5m. The model is set on two floors, a
ground floor and a first floor, thus enabling the research to test for two distinct zones. Within the simulation tool
setup, and for all the simulations carried out, no mechanical heating or cooling system was specified, therefore

6
 the conditions within the building were free floating. Also the building was specified as being a generic office
160 space with occupancy density set using a good practice design standard of 0.1 persons/m2 (see ASHRAE [55]).
Varied building occupancy, defined as a fraction of the entire possible occupancy based on the available area was
used between 0700 and 1800 hours. The glazing ratio albeit rotated along the four cardinal directions (North,
South, West and East) was kept constant at 40% of the façade. Whereas the 40% may appear as excessive,
this is not uncommon in office buildings as often the highly transparent nature of a building façade offers the
165 type of architectural signature most architects tend to assign their buildings to ensure adequate daylighting and
interaction with the outside [56]. Also, no shading was assigned as this typically has to be orientation sensitive.
This would have rendered the building model quite specific in terms of its design. Rather solar heat gain control,
or indeed its absence, was factored for by making use of glazing units with varying solar heat gain coefficients, g.
The outdoor climate was set for that typical of the Maltese Islands, representative of a Mediterranean climate
170 with Koppen Climate Classification Csa [57] with an annual average high temperature of 22.3o C and an average
low of 14.9o C [58]. The parameters studied include those typically used to test the passive behaviour of a
building, and include wall and roof U-values, ventilation rates, glazing specifications and building orientation,
with the latter in this case referring to location of the glazed opening. For each of the parameters investigated
three or more variables were chosen as shown in Table 1. For each scenario a set of 5 parameters made up of
175 one of the listed variables was modelled, with the dataset containing the hourly temperature values for an entire
year extracted to be analysed later. Based on the number of parameters and variables studied, a total of 324
combinations/scenarios were modelled for each floor. For each case, the annual hourly indoor temperature was
calculated for each of the floors.

Figure 2: Building 1 geometrical model having a square plan layout (10m × 10m) with an internal height per floor of 3.5m and
having two floors.

2.3.2. Building number 2

180 In order to further test the Algorithm a second test building was also modelled. This time the building was
modelled to have a short 5m and a long 10m side, but other than that, all other base assumptions (occupancy
density, glazing ratio, etc.) utilised for the first building were utilised in this second building as well. Also,
in this case the number of parameters was increased to 7 to take into account the variation in internal height
and the window location, that is if it is placed on the shorter or on the long side. Figure 3 shows the second
185 modelled building, with the glazing placed on either the short or the long side.
Similar to the first building case a number of parameters were chosen to study the free floating performance

7
 Table 1: Parameters and Variables Investigated for the building 1 case.

Parameter Number of Variables Detailed Description

Variables
Wall U-value 3 1.6 Similar building materials utilised (cast concrete in-
−2 −1
(W m K ) ternally plastered), with inbuilt insulation modified
to reach the required U-value
1
0.5
Roof U-value 3 0.5 Similar building materials utilised (cast concrete in-
(W m−2 K −1 ) ternally plastered), with inbuilt insulation modified
to reach the required U-value
0.3
0.2
Glazing Orientation 4 North Glazing orientation modified to coincide with all four
cardinal directions
South
East
West
Glazing Details 3 6.12/0.81 Single clear glazing
U-value/g-value 3.23/0.76 Double clear glazing 6-12-6 Air
−2 −1
(W m K /unitless) 2.31/0.43 Double clear Low-e glazing 6-12-6 Air
Air Changes per Hour 3 2 A ventilation rate was imposed to mimic natural ven-
tilation ingress into the modelled spaces. Although
this can be considered as a simplification of the prob-
lem, using this method ensures that the ventilation
rate can be specified and controlled as a variable.
Using more complex tools would have resulted in
having to control other design variables, with the
ventilation being an output result rather than a con-
trollable variable, as was required for this analysis.
A similar approach, where the ventilation rate was
fixed, was used by Gong et al. [42]
5
7

8
 (a) Building 2 short side (b) Building 2 long side

Figure 3: The second modelled building with the glazing on the short and long side respectively.

of the building. For each parameter a number of variables were chosen and modelled, thus creating 384 unique
combinations for which the annual hourly indoor temperature was calculated for each floor. Table 2 summarises
the parameters and variable investigated.

190 2.4. The big bang big crunch algorithm

In this section, the BB-BC optimisation methodology used throughout this study is explained in five steps.

Step 1 - The initial population (Big-Bang phase)

An initial set of solutions, referred to as candidates forming part of the first population, is created randomly
and distributed uniformly across the search space. For this initial population only, a uniform distribution is used
195 in creating the initial set of solutions to eliminate any candidates lying beyond the limits of the search space.
In addition, using a uniform distribution avoids giving any preference to any particular location in the search
space. The population size (N ) is kept constant throughout subsequent generations in a particular optimisation
run.

Step 2 - Evaluation of the centre of mass (Big-Crunch phase)

200 The fitness value (fi ) of each candidate (i) pertaining to the population defined in the previous step is
evaluated according to the predefined fitness (objective) function that is to minimise the number of hours
outside the comfort range as detailed in section 2.5. This fitness value is used to calculate the centre of mass
(b)
DVc for each design variable (DV b ) according to equation 1 for a minimisation problem.

DV (b)
PN
i=1 fi
DVc(b) = PN 1 . . . b = 1, 2, 3, . . . , NDV (1)
i=1 fi

where NDV is the number of design variables set during the optimisation problem definition. In addition,
205 the value of the design variables of the candidate with the lowest fitness value in the population are recorded
(b)
and denoted as DVbest .

Step 3 - Generation of a new population

In this phase, a new population is generated based on the results obtained from Step 2. The new design
(b)
variables DVnew are generated randomly similarly to Step 1 but in this case, the standard normal distribution

9
 Table 2: Parameters and Variables Investigated for the building 2 case.

Parameter Number of Variables Detailed Description

Variables
Wall U-value 2 1.6 Similar building materials utilised (cast concrete in-
(W m−2 K −1 ) ternally plastered), with inbuilt insulation modified
to reach the required U-value
0.5
Roof U-value 2 0.5 Similar building materials utilised (cast concrete in-
(W m−2 K −1 ) ternally plastered), with inbuilt insulation modified
to reach the required U-value
0.2
Window Location 2 Short Side Window located on short side
Long Side Window located on long side
Internal Height (m) 3 2.6 Floor to ceiling height
3
4
Glazing Orientation 4 North Glazing orientation modified to coincide with all four
cardinal directions
South
East
West
Glazing Details 2 3.23/0.76 Double clear glazing 6-12-6 Air
U-value/g-value
(W m−2 K −1 /unitless) 2.31/0.43 Double clear Low-e glazing 6-12-6 Air
Air Changes per Hour 2 2 A ventilation rate was imposed to mimic natural ven-
tilation ingress into the modelled spaces. Although
this can be considered as a simplification of the prob-
lem, using this method ensures that the ventilation
rate can be specified and controlled as a variable.
Using more complex tools would have resulted in
having to control other design variables, with the
ventilation being an output result rather than a con-
trollable variable, as was required for this analysis.
A similar approach, where the ventilation rate was
fixed, was used by Gong et al. [42]
8

10
210 N (0, 1)(b) is used in the generation of the new design variable as indicated equation 2

(b)
(b) (b) αN (0, 1)(b) nDV
DVnew = β × DVc(b) + (1 − β) × DVbest + . . . b = 1, 2, 3, . . . , NDV (2)
I
(b)
Where the β parameter controls the effect of the previous best solution (DVbest ) and can vary from [0, 1].
(b)
DVc is the centre of mass on the newly created candidates [30]. The α parameter falls in the range (0, 1] and
controls the extent of the search space. I is the generation number indicating the amount of populations created
(b)
whilst nDV is the total amount of discrete values that the design variable (b) can have. Note that DVnew should
(b)
215 be rounded to the nearest integer DVr new and limited within the search space where:

(b) (b) (b)


 DVmin , if DVnew < DVmin

DVr(b)
new =
(b) (b) (b) (b)
round(DVnew ), if DVmax ≥ DVnew ≥ DVmin · · · · · · b = 1, 2, 3, . . . , NDV (3)

(b) (b) (b)


 DVmax , if DVnew > DVmax

(b) (b)
DVmin and DVmax are the minimum and maximum design variable limits.

Step 4 - Convergence
Repetition of steps 2 and 3 should be carried out until the algorithm converges. In this study, when the best
candidate reoccurs over a definite amount of consecutive generations, the algorithm is assumed to be converged
220 and the best candidate is considered as the optimal solution in the given search space.

2.5. Objective function

As in any optimisation methodology, the BB-BC method described above requires an objective function
φ(xi ) where xi is the ith design variable which needs to be optimised by determining local maxima and minima
∂φ
defined by ∂xi = 0 and then the global maximum or minimum of the problem. In the passive design of buildings
225 various objective functions can be utilised depending on the principal aim of the designer. One such common
parameter is the number of hours during one single year that the temperatures in a particular zone fall outside
a predefined limit i.e. Tmin ≤ Tζ ≤ Tmax , where ζ refers to a particular zone in the building. Tmin and Tmax
can be established on the basis of desired temperature ranges depending on say thermal comfort of occupants.
Therefore the objective is to minimise the number of hours which are outside the comfort zone, φ.
230 As already mentioned, the adaptive thermal comfort model established under the European Standard EN
15251 [49] is used. The reason for this choice is the widespread use of adaptive thermal comfort models which
are specifically aimed at passive buildings not employing Heating, Mechanical ventilation and Air-Conditioning
but can have natural ventilation due to say openable windows. In this model, for acceptable comfort conditions,
Tζ must fall in the range [Tmin (Trm ), Tmax (Trm )] where Trm is an exponentially-weighted running mean of the
235 exterior temperature as follows:

Ny −1
X
Trm = (1 − Ψ) ΨTe,i (4)
i=0

Where Ψ is a constant recommended in EN15251 [49] to be 0.8, Ny is the number of days in a year and Te,i
is the mean external temperature of the previous day. The standard groups different thermal comfort levels in
three categories. In order to define these, maximum and minimum lines are used to define a region for each

11
 category. The larger the region, the greater the tolerance to temperature extremes. The lines defining the
240 categories are defined mathematically as follows:

Tmax/min = 0.33Trm + 18.8 ± ∆T (5)

Where ∆T = 2◦ C for Category I, ∆T = 3◦ C for Category II and ∆T = 4◦ C for Category III.

These limits apply for 10◦ C < Trm < 30◦ C for the upper limit and 10◦ C < Trm < 30◦ C for the lower limit.
If the exterior running mean temperatures is less than 15◦ C then the lower limits of the acceptable temperatures
remain constant at 21◦ C (category I), 20◦ C (category II), and 19◦ C (category III) respectively.

245 3. Results and discussion

3.1. Building energy simulation results

The building energy model is used to run all simulation for different combinations of design variables for
both building 1 (324 simulations) and building 2 (384 simulations). This enumeration search enabled validation
of the BB-BC code.
250 For each combination of design variables, a scatter plot of interior temperature against the running mean
temperature can be plotted and the number of points which fall within particular categories visualised. The
optimal combination of design variables which minimises the points which fall outside the limits are shown in
Figure 4 for buildings 1 and 2 and for each of the two floors modelled. It must be noted that the optimal
combination of the design variables for say Category I is not necessarily the same as the optimal combination
255 for Category III comfort. In this section, to illustrate the resulting scatter plots, the optimal solution for a
Category III building is presented.
It can be observed that the ground floor for building 1 performs very poorly especially when the running
mean temperature is below 20◦ C. The first floor performs much better within the category II and III range.
Still, below a running mean temperature of 20◦ C the building performs poorly even under optimal design
260 variable conditions. Building 2 shows better overall performance but again the floor has more points falling
within the comfort region delineated by category II and III.

3.2. Optimisation of α and β

The optimal solution obtained from the BB-BC algorithm can be evaluated for its correctness by comparing
the minimum number of hours as calculated from the enumeration search. The percentage Success Rate (SR)
265 is defined as follows:

BB-BC algorithm succesful runs

%SR = × 100% (6)
Ntrials
Where Ntrials is the number of trials. Using Ntrials = 10, 000 as the number of trials of the algorithm for
each α and β in the range [0, 1]. It transpires that α should always be set to 1 for the best SR. On the other
hand it was found that the results for varying β are not so trivial. For this reason, the SR for varying β and
for both building floors with category I, II and III targets are shown in Figures 5 and 6 for buildings 1 and 2
270 respectively. For each of the cases, the population size N used for the algorithm is changed for N = {15, 20, 35}.
The objective here is to find the ideal value for β which gives the highest SR.

12
 Interior Temperature (°C)

Interior Temperature (°C)

35 35
CategoryI CategoryI
CategoryII CategoryII
30 CategoryIII 30 CategoryIII

25 25

20 20

15 15

10 10
10 15 20 25 30 10 15 20 25 30
Running mean temperature (°C) Running mean temperature (°C)

(a) Ground floor, building 1 (b) First floor, building 1

Interior Temperature (°C)

Interior Temperature (°C)

35 35
CategoryI CategoryI
CategoryII CategoryII
30 CategoryIII 30 CategoryIII

25 25

20 20

15 15

10 10
10 15 20 25 30 10 15 20 25 30
Running mean temperature (°C) Running mean temperature (°C)

(c) Ground floor, building 2 (d) First floor, building 2

Figure 4: Hourly temperature scatter for the optimal solution as calculated from the model for category III.

13
 For building 1, the variation of the SR with β is not very sensitive. For the ground floor, the optimal β
is 0 for all categories with success rates exceeding 85%. On the other hand, for the first floor, the optimum β
is in the range 0.6 ≤ β ≤ 0.75 but smaller β values will not have a substantial impact on the SR (less than
275 5% decrease). The β parameter affects the datum from which new populations are generated. When β is set
to unity, the new population is created around the center of mass calculated from step 2 in section 3.4 whilst
when β = 0, the new population will be centered around the best fit solution of the previous population. As
illustrated in Figs. 5 and 6, choosing a value for β at the extremes has an adverse effect of the SR. This effect
stems from the particular surface of the objective function to be optimized which is obviously unknown to the
280 researcher but usually it is characterized by a complex pattern of local minima and maxima. How this optimum
in the β varies with different building zones is a non-trivial issue and possibly the best way to investigate this
is through a brute force approach. This is outside of the scope of this paper and is not envisaged to have an
important impact on the overall choice of β. Beyond β = 0.8, the SR starts deteriorating significantly falling
to less than 40% as β → 1.
285 For building 2, the situation changes in that the optimum for all floors and categories is found in the range
0.85 ≤ β ≤ 0.95. This suggests that different building design variables can alter the choice of β. In any case
choosing β = 0.8 does not degrade the SR substantially in both the building 1 and building 2 cases. Choosing a
value which is too small will degrade substantially the success rate for this building 2 case. Therefore, considering
the two situations investigated here, β = 0.8 gives the best compromise.

290 3.3. Effect of population size

The choice of the population size N used in the BB-BC algorithm is important because too low a value
can cause the success of the algorithm to be low. This trend is the result of insufficient candidates to explore
adequately the design space under investigation thus the algorithm converges prematurely to a local minimum.
On the other hand choosing a high sample size causes a substantial increase in computational cost of increased
295 number simulations. Figure 7 shows the SR variation with the sample size for the different categories. The
latter has minimal effect on the SR. For building 1 having 5 design variables, N > 13 gives a SR higher than
70%. For building 2 having 7 design variable, a SR greater than 70% occurs at N > 20. The ratio of the
population size to the number of design variables is in between 2.6 to 2.9 for a SR which is greater than 70%.
As indicated in section 2.4, the BB-BC algorithm is a non-deterministic optimisation approach therefore
300 consecutive runs might converge to different solutions. In view of this intrinsic shortcoming, the SR as defined
by equation 6 can be used to identify the level of confidence in the results obtained via the BB-BC algorithm.
In other words, a high SR ratio means that a lower amount of consecutive runs are required to obtain the global
optimum thus increasing the algorithm efficiency.

3.4. Error analysis for ’failed’ results

305 The SR discussed under section 3.2 only gives an indication of whether the BB-BC algorithm succeeds in
finding the exact optimum as found from an enumeration search. In practice, even if the algorithm technically
fails to obtain the global optimum result, the difference between the calculated one and the actual optimum
might still be, within practical reasons, close to the optimum. Therefore it is also useful to know the error
between a particular BB-BC trial and the true optimum. Out of a total of 10,000 trials such an error can be

14
 100 100
Success rate (%)

Success rate (%)

90 90
80
80
70
70
60
60
50
50
40 N=15 N=15
30 N=20 40 N=20
N=35 N=35
20 30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) Ground floor, building 1, category I (b) First floor, building 1, category I

100 100
Success rate (%)

Success rate (%)

90 90

80 80

70 70

60 60

50 50
N=15 N=15
40 N=20 40 N=20
N=35 N=35
30 30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(c) Ground floor, building 1, category II (d) First floor, building 1, category II

100 100
Success rate (%)

Success rate (%)

90
90
80
80
70
60 70
50
60
40 N=15 N=15
N=20 50 N=20
30
N=35 N=35
20 40
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(e) Ground floor, building 1, category III (f) First floor, building 1, category III

Figure 5: Success rate variation with β for the 10,000 runs of the BB-BC algorithm. Results are shown here for building 1 for each
category and each floor level.

15
 70 90
Success rate (%)

Success rate (%)

60 80
50
70
40
60
30
50
20
N=15 N=15
10 N=20 40 N=20
N=35 N=35
0 30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) Ground floor, building 2, category I (b) First floor, building 2, category I

70 90
Success rate (%)

Success rate (%)

60 80
50
70
40
60
30
50
20
N=15 N=15
10 N=20 40 N=20
N=35 N=35
0 30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(c) Ground floor, building 2, category II (d) First floor, building 2, category II

70 90
Success rate (%)

Success rate (%)

60 80

50 70

40 60

30 50
N=15 N=15
20 N=20 40 N=20
N=35 N=35
10 30
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(e) Ground floor, building 2, category III (f) First floor, building 2, category III

Figure 6: Success rate variation with β for the 10,000 runs of the BB-BC algorithm. Results are shown here for building 2 for each
category and each floor level.

16
 100 100
90 90
Success Rate (%)

Success Rate (%)

80 80
70 70
60 60
50 50
40 CategoryI 40 CategoryI
30 CategoryII 30 CategoryII
CategoryIII CategoryIII
20 20
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
N N

(a) Ground floor, building 1 (b) First floor, building 1

100 100
90 90
Success Rate (%)

Success Rate (%)

80 80
70 70
60 60
50 50
40 CategoryI 40 CategoryI
30 CategoryII 30 CategoryII
CategoryIII CategoryIII
20 20
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
N N

(c) Ground floor, building 2 (d) First floor, building 2

Figure 7: Success rate against the population size, N .

17
 6 25

Percentage Error (%)

Percentage Error (%)

CategoryI CategoryI
5 CategoryII CategoryII
CategoryIII 20 CategoryIII
4
15
3
10
2

5
1

0 0
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
N/(number of design variables) N/(number of design variables)

(a) Ground floor, building 1 (b) First floor, building 1

50 70
Percentage Error (%)

Percentage Error (%)

CategoryI CategoryI
CategoryII 60 CategoryII
40 CategoryIII CategoryIII
50
30
40
20
30

10
20

0 10
0 1 2 3 4 5 6 0 1 2 3 4 5 6
N/(number of design variables) N/(number of design variables)

(c) Ground floor, building 2 (d) First floor, building 2

Figure 8: Percentage error against the ratio of the population size, N to the number of design variables.

310 averaged out. This error is mathematically defined as follows

P P
i (φi − φ0 )
= = (7)
Ntrials Ntrials
where φi is the solution for trial i, φ0 is the true solution and i is the error between the trial and true
solution. The percentage  is plotted against the population size divided by the number of design variables for
both building 1 and building 2 in Figure 8. The category class is also plotted in each figure. Category III in
fact presents the highest error margins and at very low sample number to design variable ratios the errors can
315 exceed 40% which is unacceptable. Considering a ratio of 2.9 which provides success rates greater than 70%,
the error found in the unsuccessful results is less than 11% for building 1 and less than 34% for building 2 when
considering category III. This latter result could be concerning for the designer and in such cases it would be
recommended to run the algorithm multiple times. When designing category I buildings, the error is less than
12% for both buildings considered.
320 Further to the results and discussion explained here, it is also interesting to plot results for SR vs . These
plots are shown in Figure 9. This can provide guidance on what to expect for situations giving a low success
rate compared to the resulting error. In general, for all zones and for both buildings,  increases monotonically
with decreasing SR. On each line plot, the different points on the lines refer to a different sample size. For
convenience, three sample sizes are shown on each line. These plots further highlight the importance of choosing
325 the right value for N . The worst case result is found for category III class with errors exceeding 60% for SR
which is around 30% when using a sample size of 5. When using N = 15, the error is minimised to 35% with

18
 N=40 N=40 N=40 N=40
100
N=15 100
CategoryI CategoryI
90 N=15 90

Success Rate (%)

Success Rate (%)

CategoryII N=15N=15 N=15 CategoryII
80 CategoryIII 80 CategoryIII

70 70
60 60
50 N=5 50
40 N=5 40
N=5 N=5 N=5 N=5
30 30
20 20
0 1 2 3 4 5 6 0 5 10 15 20 25
Percentage Error (%) Percentage Error (%)

(a) Ground floor, building 1 (b) First floor, building 1

80 100
N=40 CategoryI N=40 CategoryI
70 90 N=40
Success Rate (%)

Success Rate (%)

N=40 CategoryII CategoryII
N=40 CategoryIII CategoryIII
60 80
N=15 N=15
50 70 N=15 N=15

40 60
N=15 N=15
30 50
20 N=5 40
N=5 N=5 N=5
10 30
N=5 N=5
0 20
0 5 10 15 20 25 30 35 40 45 50 10 20 30 40 50 60 70
Percentage Error (%) Percentage Error (%)

(c) Ground floor, building 2 (d) First floor, building 2

Figure 9: Success rate against percentage error. Different sample sizes N are shown.

success rates of 70%.

4. Computational Efficiency

The presented optimisation of the BB-BC algorithm required validation against an enumeration search
330 approach where all combinations of design variables are simulated using the building energy modelling tool. In
practice, the scope of using a BB-BC algorithm is to avoid using a brute force enumeration search and therefore
substantially reduce the computational time required. As a measure of the computational savings of using the
BB-BC algorithm as opposed to the enumeration search, we define the ratio Θ as follows:

Number of simulations with BB-BC algorithm I ×N

Θ=1− =1− Y (8)
Total number of design variable combinations nDV
∀DV

This formula implies that if Θ = 1 we have a 100% reduction in the number of simulations, thus time savings,
335 and if Θ = 0 it means that there are no computational and time savings over the brute force enumeration
scheme. Using the results found for the two case studies presented, the number of iterations for convergence
can be taken as I = 7 while the population size in relation to the number of design variables NDV can be taken
as approximately N ≈ 2.9NDV we have:

7 × 2.9 × NDV 20.3 × NDV

Θ=1− Y =1− Y (9)
nDV nDV
∀DV ∀DV

19
 To have a first order estimate of the computational savings we assume that the number of discrete values of
Y
340 DV is the same for each design variable i.e. nDV = nNDV . This gives
∀DV

20.3
Θ=1− (10)
nNDV
This shows that as the number of design variables and the number of discrete values tested for each design
variable increases, efficiency of using the BB-BC method increases at a very fast rate. When the number of
design variables is 5 or more, savings of more than 90% are expected.

5. Discussion

345 While the aims of the paper focus on the fine-tuning of the BB-BC approach in building simulation specifically
in relation to the passive design of buildings, the physical aspects of the results should be discussed in order to
substantiate further the need for optimisation techniques. In addition this helps to link the adoption of such
an algorithm to the physical processes involved.
Table 3 gives the design variable combinations which give minimum hours outside of the comfort range
350 within the category III region for both floors of buildings 1 and 2. The preferred orientation for both buildings
and their respective floors is north which ensures that no direct solar radiation penetrates through the windows.
This result is to be expected. Nonetheless, other variables such as Glazing U-value, should be high to minimise
the number of hours outside of the comfort temperature range in the case of building 1 and vice-versa for the
case of building 2. The reason for this is the rectangular shape of building number 2 since window to wall ratio
355 is maintained the same between both buildings. The same diametrically opposed results are encountered with
the ACH, the wall and roof U-value. The reasons for these different results in these design variables is clearly
entrenched in the shape of the building with the square shaped building requiring less insulation. It must be
emphasised that these results are obtained for this particular climate zone. Certain conclusions such as this
may be difficult to arrive at without adopting a multi-design variable approach and considering the various
360 combinations of these design variables. Usual notions of building design may fail if there is no consideration
of the various parameter combinations which are possible. The BB-BC method has been shown to provide an
efficient way to test situations with a large number of design variables.

Table 3: Combinations of design variables giving the minimal number of hours outside the category III range

Building Floor Internal Location Orientation Glazing ACH Walls Roof Hrs
Height of win- [◦ ] U-value U-value U-value outside
2 2 2
[m] dow [W/m K] [W/m K] [W/m K] range
1 ground Not a Not a 0.00 6.12 7.00 1.60 0.49 4751
variable variable
1 first Not a Not a 0.00 6.12 7.00 1.60 0.49 5809
variable variable
2 ground 2.30 short side 0.00 2.31 2.00 0.50 0.20 7902
2 first 2.60 short side 0.00 2.31 2.00 0.50 0.20 4835

20
 6. Conclusions

This paper has presented the use of the BB-BC optimization approach for use in the passive design of
365 buildings, specifically the assessment of the number of hours annually which would result in uncomfortable
thermal comfort conditions inside buildings, as defined by EN 15251.
Apart from the optimization exercise in itself, the paper also looks at what specific parameters, that is, α
and β should be used to obtain the best results, when using the BB-BC optimization method for such a case
scenario. In fact it was found that α should be 1, whilst β should in the vicinity of 0.8.
370 For the buildings analysed in this research, the ratio of the sample size to the number of design variables
should be between 2.6 and 2.9 for BB-BC success rates higher than 70%. The BB-BC algorithm applied to
Category III designs can yield more errors in the calculation, whilst errors for failed runs of the BB-BC algorithm
show errors which are less than 12% for Category I buildings.
The computational efficiency of using the BB-BC algorithm over an enumeration search increases at a very
375 fast rate when either the number of design variables or the number of discrete values of the design variable
increases. For 5 or more building design variables, savings of more than 90% are to be expected.
From a more practical perspective, it has been shown that for the climate considered, the choice of U-values
might be non-trivial when minimising the number of hours which are outside the thermal comfort zone with a
rectangular shaped building requiring more insulation than a square shaped building.
380 The applicability of the conclusions provided here over a wide range of building configurations and designs
is still an open question which needs to be tackled on the basis of the outcomes of this work. In addition,
the generality of these results for different objective functions other than comfort temperatures as defined in
EN 15251 is not addressed in this work and is certainly a research question which remains to be answered.
Furthermore, the efficiency of the BB-BC algorithm with respect to other similar algorithms should be the
385 scope of future studies.

References

[1] A. Prieto, U. Knaack, T. Auer, T. Klein, Passive cooling & climate responsive façade design: Exploring the
limits of passive cooling strategies to improve the performance of commercial buildings in warm climates,
Energy and Buildings 175 (2018) 30–47. doi:[Link]
390 URL [Link]

[2] X. Sun, Z. Gou, S. S.-Y. Lau, Cost-effectiveness of active and passive design strategies for existing building
retrofits in tropical climate: Case study of a zero energy building, Journal of Cleaner Production 183 (2018)
35–45. doi:[Link]
URL [Link]

395 [3] X. Wang, X. Mai, B. Lei, H. Bi, B. Zhao, G. Mao, Collaborative optimization between passive design
measures and active heating systems for building heating in Qinghai-Tibet plateau of China, Renewable
Energy 147 (2020) 683–694. doi:[Link]
URL [Link]

[4] X. Chen, J. Huang, H. Yang, J. Peng, Approaching low-energy high-rise building by integrating passive
400 architectural design with photovoltaic application, Journal of Cleaner Production 220 (2019) 313–330.

21
 doi:[Link]
URL [Link]

[5] P. Capros, G. Zazias, S. Evangelopoulou, M. Kannavou, T. Fotiou, P. Siskos, A. D. Vita, K. Sakellaris,

Energy-system modelling of the EU strategy towards climate-neutrality, Energy Policy 134 (2019) 110960.
405 doi:[Link]
URL [Link]

[6] É. Mata, A. S. Kalagasidis, F. Johnsson, Contributions of building retrofitting in five member states to EU
targets for energy savings, Renewable and Sustainable Energy Reviews 93 (2018) 759–774. doi:https:
//[Link]/10.1016/[Link].2018.05.014.
410 URL [Link]

[7] K. Biswas, T. Patel, S. Shrestha, D. Smith, A. Desjarlais, Whole building retrofit using vacuum insulation
panels and energy performance analysis, Energy and Buildings 203 (2019) 109430. doi:[Link]
10.1016/[Link].2019.109430.
URL [Link]

415 [8] B. Rosti, A. Omidvar, N. Monghasemi, Optimum position and distribution of insulation layers for exterior
walls of a building conditioned by earth-air heat exchanger, Applied Thermal Engineering 163 (2019)
114362. doi:[Link]
URL [Link]

[9] H. Huang, Y. Zhou, R. Huang, H. Wu, Y. Sun, G. Huang, T. Xu, Optimum insulation thicknesses and
420 energy conservation of building thermal insulation materials in Chinese zone of humid subtropical climate,
Sustainable Cities and Society 52 (2020) 101840. doi:[Link]
URL [Link]

[10] J. Morrissey, T. Moore, R. E. Horne, Affordable passive solar design in a temperate climate: An experiment
in residential building orientation, Renewable Energy 36 (2) (2011) 568–577. doi:[Link]
425 1016/[Link].2010.08.013.
URL [Link]

[11] A. Albatayneh, D. Alterman, A. Page, B. Moghtaderi, The Significance of the Orientation on the Overall
buildings Thermal Performance-Case Study in Australia, Energy Procedia 152 (2018) 372–377. doi:https:
//[Link]/10.1016/[Link].2018.09.159.
430 URL [Link]

[12] M. Winkler, F. Antretter, J. Radon, Critical discussion of a shading calculation method for low energy
building and passive house design, Energy Procedia 132 (2017) 33–38. doi:[Link]
[Link].2017.09.627.
URL [Link]

435 [13] M. Bonello, D. Micallef, S. P. Borg, Humidity micro-climate characterisation in indoor environments : A
benchmark study, Journal of Building Engineering 28 (April 2019) (2020) 101013. doi:10.1016/[Link].

22
 2019.101013.
URL [Link]

[14] L. Giovannini, F. Favoino, A. Pellegrino, V. R. M. L. Verso, V. Serra, M. Zinzi, Thermochromic glazing

440 performance: From component experimental characterisation to whole building performance evaluation,
Applied Energy 251 (2019) 113335. doi:[Link]
URL [Link]

[15] T. P. Obrecht, M. Premrov, V. Ž. Leskovar, Influence of the orientation on the optimal glazing size for
passive houses in different European climates (for non-cardinal directions), Solar Energy 189 (2019) 15–25.
445 doi:[Link]
URL [Link]

[16] T. Ashrafian, N. Moazzen, The impact of glazing ratio and window configuration on occupants’ comfort
and energy demand: The case study of a school building in Eskisehir, Turkey, Sustainable Cities and
Society 47 (2019) 101483. doi:[Link]
450 URL [Link]

[17] E. Cuce, F. Sher, H. Sadiq, P. M. Cuce, T. Guclu, A. B. Besir, Sustainable ventilation strategies in
buildings: CFD research, Sustainable Energy Technologies and Assessments 36 (2019) 100540. doi:https:
//[Link]/10.1016/[Link].2019.100540.
URL [Link]

455 [18] D. Micallef, V. Buhagiar, S. P. Borg, Cross-ventilation of a room in a courtyard building, Energy and
Buildings 133 (2016) 658–669. doi:10.1016/[Link].2016.09.053.
URL [Link]

[19] J. Nievergelt, Exhaustive Search, Combinatorial Optimization and Enumeration: Exploring the Potential
of Raw Computing Power, in: V. Hlaváč, K. G. Jeffery, J. Wiedermann (Eds.), SOFSEM 2000: Theory
460 and Practice of Informatics, Springer Berlin Heidelberg, Berlin, Heidelberg, 2000, pp. 18–35.

[20] S. S. Rao, Engineering optimization: theory and practice, 4th Edition, John Wiley & Sons, Inc., 2009, Ch.
13 Modern Methods of Optimization, p. 693.

[21] H. JH., Adaptation in natural and artificial systems, MIT Press, 1992.

[22] R. L. Haupt, S. E. Haupt, Practical Genetic Algorithms, 2nd Edition, John Wiley & Sons, Inc., 2003.

465 [23] S. Kirkpatrick, C. D. Gelatt, M. P. Vecchi, Optimization by Simulated Annealing, Science 220 (4598) (1983)
671–680. doi:10.1126/science.220.4598.671.
URL [Link]

[24] J. Kennedy, R. Eberhart, Particle swarm optimization, in: Proceedings of ICNN’95 - International Con-
ference on Neural Networks, Vol. 4, 1995, pp. 1942–1948 vol.4. doi:10.1109/ICNN.1995.488968.

470 [25] A. Colorni, M. Dorigo, V. Maniezzo, Distributed Optimization by Ant Colonies, in: Proceedings of the
First European Conference on Artificial Life, 1991.

23
 [26] M. Dorigo, V. Maniezzo, A. Colorni, Ant system: optimization by a colony of cooperating agents, IEEE
Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics) 26 (1) (1996) 29–41. doi:10.1109/
3477.484436.

475 [27] O. K. Erol, I. Eksin, A new optimization method: Big Bang-Big Crunch, Advances in Engineering Software
37 (2) (2006) 106–111. doi:[Link]
URL [Link]

[28] A. Kaveh, S. Talatahari, Size optimization of space trusses using Big Bang-Big Crunch algorithm, Comput-
ers & Structures 87 (17) (2009) 1129–1140. doi:[Link]
480 URL [Link]

[29] A. Kaveh, S. Talatahari, A discrete Big Bang - Big Crunch algorithm for optimal design of skeletal
structures, Asian Journal of Civil Engineering 11 (1) (2010) 103–122.
URL [Link]
40{&}md5=ce293ff10b33f57bf4bb2929a2e04493

485 [30] C. V. Camp, Design of Space Trusses Using Big Bang - Big Crunch Optimization 133 (July) (2007) 999–
1008. doi:10.1061/(ASCE)0733-9445(2007)133.

[31] O. Hasanebi, S. K. Azad, Reformulations of Big Bang-Big Crunch Algorithm for Discrete Structural Design
Optimization (feb 2013). doi:10.5281/zenodo.1085876.
URL [Link]

490 [32] H. Tang, J. Zhou, S. Xue, L. Xie, Big Bang-Big Crunch optimization for parameter estimation in structural
systems, Mechanical Systems and Signal Processing 24 (8) (2010) 2888–2897. doi:[Link]
1016/[Link].2010.03.012.
URL [Link]

[33] Y. Labbi, D. B. Attous, Big Bang-Big Crunch optimization algorithm for economic dispatch with
495 valve-point effect, Journal of Theoretical and Applied Information Technology 16 (1) (2010) 48–56.
URL [Link]
40{&}md5=a4629edc4cc0e75b6c24950f999acd37

[34] P. Y. Tabakov, Big bang - big crunch optimization method in optimum design of complex composite
laminates, Int. J. Mechan. Aero. Ind. Mech. Manufact. Eng. 5 (5) (2011) 927–931.

500 [35] S. Kazemzadeh Azad, T. Ak\i\cs, Automated selection of optimal material for pressurized multi-layer
composite tubes based on an evolutionary approach, Neural Computing and Applications 29 (7) (2018)
405–416. doi:10.1007/s00521-016-2563-6.
URL [Link]

[36] T. Kumbasar, b. Eksin, M. Güzelkaya, E. Ye\csil, Big Bang Big Crunch Optimization Method Based Fuzzy
505 Model Inversion, in: A. Gelbukh, E. F. Morales (Eds.), MICAI 2008: Advances in Artificial Intelligence,
Springer Berlin Heidelberg, Berlin, Heidelberg, 2008, pp. 732–740.

24
 [37] H. M. Gen, A. K. Hocaoglu, Bearing-Only Target Tracking Based on Big Bang Big Crunch Algorithm,
in: 2008 The Third International Multi-Conference on Computing in the Global Information Technology
(iccgi 2008), 2008, pp. 229–233. doi:10.1109/ICCGI.2008.53.

510 [38] X. Chen, H. Yang, A multi-stage optimization of passively designed high-rise residential buildings in mul-
tiple building operation scenarios, Applied Energy 206 (2017) 541–557. doi:[Link]
[Link].2017.08.204.
URL [Link]

[39] X. Chen, H. Yang, W. Zhang, Simulation-based approach to optimize passively designed buildings: A
515 case study on a typical architectural form in hot and humid climates, Renewable and Sustainable Energy
Reviews 82 (2018) 1712–1725. doi:[Link]
URL [Link]

[40] S. Gou, V. M. Nik, J.-L. Scartezzini, Q. Zhao, Z. Li, Passive design optimization of newly-built residential
buildings in Shanghai for improving indoor thermal comfort while reducing building energy demand, Energy
520 and Buildings 169 (2018) 484–506. doi:[Link]
URL [Link]

[41] A. Mukhtar, M. Z. Yusoff, K. C. Ng, The potential influence of building optimization and passive design
strategies on natural ventilation systems in underground buildings: The state of the art, Tunnelling and
Underground Space Technology 92 (2019) 103065. doi:[Link]
525 URL [Link]

[42] X. Gong, Y. Akashi, D. Sumiyoshi, Optimization of passive design measures for residential buildings in
different Chinese areas, Building and Environment 58 (2012) 46–57. doi:[Link]
buildenv.2012.06.014.
URL [Link]

530 [43] A.-T. Nguyen, S. Reiter, P. Rigo, A review on simulation-based optimization methods applied to build-
ing performance analysis, Applied Energy 113 (2014) 1043–1058. doi:[Link]
apenergy.2013.08.061.
URL [Link]

[44] E. Znouda, N. Ghrab-Morcos, A. Hadj-Alouane, Optimization of Mediterranean building design using

535 genetic algorithms, Energy and Buildings 39 (2) (2007) 148–153. doi:[Link]
enbuild.2005.11.015.
URL [Link]

[45] R. Evins, P. Pointer, R. Vaidyanathan, S. Burgess, A case study exploring regulated energy use in domes-
tic buildings using design-of-experiments and multi-objective optimisation, Building and Environment 54
540 (2012) 126–136. doi:[Link]
URL [Link]

[46] W. Wang, R. Zmeureanu, H. Rivard, Applying multi-objective genetic algorithms in green building design
optimization, Building and Environment 40 (11) (2005) 1512–1525. doi:[Link]

25
 buildenv.2004.11.017.
545 URL [Link]

[47] J. A. Wright, H. A. Loosemore, R. Farmani, Optimization of building thermal design and

control by multi-criterion genetic algorithm, Energy and Buildings 34 (9) (2002) 959–972.
doi:10.1016/S0378-7788(02)00071-3.
URL [Link]
550 1016{%}2FS0378-7788{%}2802{%}2900071-3{&}partnerID=40{&}md5=1d1d8c3e0d5b7ce69333ae9911797de3

[48] A. Milajić, D. Beljaković, P. Petronijevic, Genetic algorithms for assigning tasks to construction machine
operators, Gradevinar 63 (2011) 749–755.

[49] DIN, EN15251 Indoor environmental input parameters for design and assessment of energy performance
of buildings addressing indoor air quality, thermal environment, lighting and acoustics;, Tech. rep.

555 [50] L. Berkeley, O. A. K. Ridge, M. B. Y. Ut-battelle, A. For, S. Energy, D. Or, T. In, A. N. Y. Form, O. R.
By, A. N. Y. Means, W. The, EnergyPlus Essentials.

[51] DesignBuilder, DesignBuilder User Guide, Tech. Rep. October (2009).

[52] A. S. Andelković, I. Mujan, S. Dakić, Experimental validation of a EnergyPlus model: Application of a

multi-storey naturally ventilated double skin façade, Energy and Buildings 118 (2016) 27–36. doi:https:
560 //[Link]/10.1016/[Link].2016.02.045.
URL [Link]

[53] A. Al-janabi, M. Kavgic, A. Mohammadzadeh, A. Azzouz, Comparison of EnergyPlus and IES to model a
complex university building using three scenarios: Free-floating, ideal air load system, and detailed, Journal
of Building Engineering 22 (2019) 262–280. doi:[Link]
565 URL [Link]

[54] W. Pereira, A. Bögl, T. Natschläger, Sensitivity Analysis and Validation of an EnergyPlus Model of a
House in Upper Austria, Energy Procedia 62 (2014) 472–481. doi:[Link]
2014.12.409.
URL [Link]

570 [55] Ashrae, Ashrae Handbook 2016: HVAC Systems and Equipment: SI Edition, ASHRAE Handbook of
Heating, Ventilating and Air-Conditioning Systems and Equipment SI, ASHRAE, 2016.
URL [Link]

[56] B. Bueno, J. M. Cejudo-López, A. Katsifaraki, H. R. Wilson, A systematic workflow for retrofitting office
façades with large window-to-wall ratios based on automatic control and building simulations, Building
575 and Environment 132 (2018) 104–113. doi:[Link]
URL [Link]

[57] M. C. Peel, B. L. Finlayson, T. A. McMahon, Updated world map of the Köppen-Geiger climate classifica-
tion, Hydrology and Earth System Sciences 11 (5) (2007) 1633–1644. doi:10.5194/hess-11-1633-2007.
URL [Link]

26
580 [58] C. Galdies, A. Said, L. Camilleri, M. Caruana, Climate change trends in Malta and related beliefs, concerns
and attitudes toward adaptation among Gozitan farmers, European Journal of Agronomy 74 (2016) 18–28.
doi:[Link]
URL [Link]

27
Paper Highlights

 Passive building design optimisation using the BB-BC algorithm has been demonstrated
 The BB-BC algorithm has been adapted for the first time to passive building design
 BB-BC optimisation can be as much as 90% more efficient than a full enumeration search
Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal relationships
that could have appeared to influence the work reported in this paper.

☒The authors declare the following financial interests/personal relationships which may be considered
as potential competing interests:

N/A