Evolving Illumination Design Following Genetic Strategies

Alice Plebe, Vincenzo Cutello and Mario Pavone

Department of Mathematics and Computer Science, University of Catania,

V.le A. Doria 6, I-95125 Catania, Italy

Keywords:

Lighting Design, Multi-objective Optimization, Genetic Algorithm, Blender, NSGA-II.

Abstract:

Interior lighting design is a challenging task where are involved multiple constraints that need to be optimized

for producing an accurate illumination avoiding possible glare. This paper, then, takes up the issue of provid-

ing a computational tool able to produce a proper lighting plan in interior spaces for a comfortable and optimal

vision in all environments, taking also into account the energy consumption as little as possible. For ﬁnding

acceptable solutions we have used the metaphor of the genetic evolution in a multi-objective model, where in-

dividuals are lists of possible light sources, their positions and lighting levels. For ﬁnding acceptable solutions

we have used the metaphor of the genetic evolution in a multi-objective model, where every individual is a list

of light sources; their positions; and lighting levels. Further, for properly evaluating each individual, we have

developed two conﬂicting objective functions, one for optimizing the level of brightness, and the second one

for maximising the energy saving, satisfying, obviously, the additional constraints to respect the architectural

structure to be lighted. From the randomly initial population of individuals generations are constructed using

crossover and mutation operators, whilst the ﬁttest offspring is preserved via an elitist Pareto-dominance se-

lection approach. In addition to the multi-objective genetic algorithm, the 3D graphic software Blender has

been used in order to reproduce the architectural space to be lighted, with the aim to evaluate then, the accu-

racy and uniformity of the produced lighting through a physical simulation of its brightness. The main goal of

the developed tool is to provide to the designer (i.e. the decision maker) a set of interiors illumination design

options, for the given environment to be lit, ensuring (i) uniform illumination distribution; (ii) accuracy of the

illumination produced; (iii) avoiding harsh brightness, and glare; and (iv) low energy consumptions. Two case

studies have been considered in our evaluation experiments, and for each of these the algorithm was performed

on two different instances and with different types of complexity respectively.

1 INTRODUCTION

Forms of artiﬁcial lighting have been introduced since

antiquity to make visual perception possible when,

and where sunlight is lacking (Wunderlich, 2003).

In most of the contemporary world a considerable

amount of time is spent indoor, and often with insuf-

ﬁcient daylight illumination. The human vision sys-

tem, like in most primates and several mammals, is

able to adapt itself to very low light levels so that we

may properly move inside interior environments, ori-

ent ourselves, and carry out actions. Unfortunately,

this adaptation occurs through the rod photoreceptor-

based scotopic vision, deprived from color and de-

tailed local feature analysis. The vision with full color

perception, and object recognition based on local fea-

tures, is produced via cone photoreceptor-based pho-

topic vision, that becomes active with adequate light

levels only (Palmer, 1999). In view of this, the light-

ing equipment selection and its placement becomes

crucial, in order to offer comfortable living environ-

ments, and improve the quality of life. Thus, integrat-

ing luminaries into the buildings aims to assign a vi-

sual perception to a sufﬁcient level for reliable recog-

nition, and cope with the environment, and, some-

times, elevate the vision to higher levels of appreci-

ation, as for instance in interior spaces hosting art-

works (Cuttle, 2015). However, providing comfort-

able and pleasant visual experiences is not an easy

task, because to determine the proper lighting equip-

ment selection, and their correct placements require

the designer to satisfy several constraints, such as the

type of occupants and the type of activities in the

given space, or the interior surface ﬁnishes, and fur-

nishings (Gordon, 2014; Livingston, 2015). Quite of-

ten, the aim to provide enough indoor light for a com-

fortable photopic vision, must take into account also

the demand in terms of energy to be spent for lighting

Plebe A., Cutello V. and Pavone M.

Evolving Illumination Design Following Genetic Strategies.

DOI: 10.5220/0006501902890296

In Proceedings of the 9th International Joint Conference on Computational Intelligence (IJCCI 2017), pages 289-296

ISBN: 978-989-758-274-5

(Sansoni et al., 2015). In U.S. the energy consumed

for lighting accounts for about 30% of the total en-

ergy consumed by commercial buildings, whilst in the

European Union the yearly consumption is over 170

TWh (Bertoldi et al., 2012).

The problem of the interior lighting design is then

a challenging task due to its complex and hard proper-

ties, and the many constraints to be satisﬁed. Indeed,

to perform a proper and accurate illumination, the de-

signer must not only take care about a right combina-

tion between daylight and artiﬁcial illumination, but

primarily must take into account: (1) the light reﬂec-

tion (specular, diffuse, or directional-diffuse), which

strongly depend on the materials composition used;

(2) degree of reﬂection; (3) shading; (4) uniform dis-

tribution of luminaries; (5) energy consumption, with

the purpose to avoid harsh brightness, and glare. The

disadvantage of using the traditional methodologies is

that the designer must specify the location, and power

of the luminaries; run a computational tool for achiev-

ing an illuminance proposal, and check if it is satis-

factory. If however it isn’t the required one, then the

computational tool must be run again.

It is clear that the interior lighting design involves

multiple factors, often conﬂicting, proving, thus, the

multi–objective properties of the problem. This pa-

per proposes a method for searching lighting solu-

tions for interior environment, based on the use of a

multi-objective genetic algorithm optimization tech-

nique, and followed by a clustering of the solutions

on the Pareto front. Typically multi-objective op-

timization methods generates too many solutions in

the ﬁnal Pareto set, and selecting a single one that

best reﬂects the preferences of the architect requires

a drastic reduction of the search space of solutions,

here implemented with clustering. The computation

of the direct illumination, necessary to compute the

ﬁtness of the population of solutions during the ge-

netic algorithm, is performed using the 3D graphic

software Blender, which allows to physically evaluat-

ing the lighting produced. Finally, results on a variety

of interior environments, with different architectural

complexities, will be shown in order to evaluate the

accuracy and efﬁciency of the presented method.

The paper is structured as follow: in Sect. 2 we

describe the evolutionary multi-objective approach,

focusing on the description and details of the algo-

rithm developed; in Sect. 3 we introduce the 3D

graphic software Blender, used as simulation environ-

ment, whilst in Sect. 4 is described the clustering of

the solutions approach, which was designed for re-

strict search space of solutions, facilitating, then, the

decision maker in choosing of the best illumination

option produced; Sect. 5 contains the experimental

results performed; and, ﬁnally, Sect. 6 contains the

concluding remarks.

2 THE NSGA-II FOR INTERIOR

ILLUMINATION DESIGN

Solve the interior illumination design problem means

satisfying multiple constraints involved but often in

conﬂict, showing a natural multi-objective approach.

Multi-objective optimization problems are character-

ized by having two or more objective functions to be

optimized; therefore, in contrast to single-objective

ones, the goal becomes to determine the set of best

tradeoffs between all the conﬂicting criteria, whose

set is called Pareto optimal set.

In this work, for solving the described problem,

we propose the well-known NSGA-II algorithm, in

which we developed a novel chromosomal represen-

tation of solutions, speciﬁcally tailored for lighting

design optimization. Each individual represents a

possible illumination conﬁguration, and it is coded

as a vector of variable length, containing a set of

lamp speciﬁcations, that is the set of features describ-

ing the luminaries in the 3D environment, including

position and orientation, intensity, color temperature

of light, and model of light ﬁxture (wall or ceiling

mount). Special operators of crossover and mutation

have been designed to handle this peculiar chromo-

somal representation. The design of such operators

is, however, facilitated by the transparency of the rep-

resentation itself. Therefore, our approach is intro-

duced especially to deal with representation of com-

plex structured individuals, and it ensures more ﬂexi-

bility with respect to previous proposals.

The Non-dominated Sorting Genetic Algorithm

II (NSGA-II) (Deb et al., 2000), is an elitist multi-

objective genetic algorithm that performs well with

real world problems, producing Pareto-optimal so-

lutions to the optimization problem. The elitist ap-

proach favours the best solutions of a population by

giving them an opportunity to be directly carried over

to the next generation. This strategy makes sure that

the best ﬁtness values do not deteriorate during the

evolution, and it enhances the probability of creating

better offspring. The elitism is integrated in the al-

gorithm by selecting the next-generation population

of size N among the best individuals from the off-

spring and the parent population combined together

(size 2N). This selection strategy, named crowded

tournament selection, takes into account two criteria:

the non-domination and the crowding distance of the

individuals. The ﬁrst is the non-domination rank of

the solution in the population, and it is used to classify

the entire 2N population into non-dominated fronts.

The second criterion is a measure of the search space

around the solution, which is not occupied by any

other solution in the population. Giving preference to

solutions that are less crowded (with larger crowding

distances) ensures a better spread among the solutions

during the evolution. These conditions make sure that

non-dominated individuals belonging to a high-rank

front and residing in a less crowded area, are selected

to reproduce more than others. The result of the al-

gorithm is the set of non-dominated solutions of the

whole ﬁnal population, namely the Pareto front.

In the implementation of the NSGA-II algorithm

proposed in this paper, at each step t of evolu-

tion, there is a population P

(t)

{

}

, which el-

ements are individuals, coding a lighting solution:

I =

, L

, ··· , L

, and L =

{

C,W

}

, p, l, w, k

The genetic code of an individual I is a ordered set of

lamp descriptions L, in which d is a code identifying

the type of commercial lamp, the second parameter

speciﬁes the type of placement: C for ceiling or W

for wall. The vector p speciﬁes the 3D coordinates

of the lamp placement, l is the intensity of the lamp

in lumen, w its electrical consumption in watt, and

k the color temperature in Kelvin degrees. Note that

the number of lamp description L in a single individ-

ual is not ﬁxed, but constrained: L

MIN

≤ L ≤ L

MAX

The initial population P

(0)

is generated randomly, us-

ing a set of predeﬁned real lamps, each with a set of

feasible combinations of intensity l and color temper-

ature k. The type of lamp d speciﬁes if the lamp can

be mounted on the ceiling or on walls. The variation

of the population is based on two fundamental opera-

tions: crossover and mutation. Given two individuals,

(1)

, ··· , L

(1)

, and I

(2)

, ··· , L

(2)

we deﬁne as two-points crossover the following func-

tion:

χ(

, I

) =



(1)

, ··· , L

(1)

, L

(2)

i+1

, ··· , L

(2)

, ··· , L

(1)

j+1

, ··· , L

(1)

(2)

, ··· , L

(2)

, L

(1)

i+1

, ··· , L

(1)

, ··· , L

(2)

j+1

, ··· , L

(2)



(1)

where i and j are two random integers such that

1 < i < j < min{L

(1)

, L

(2)

}. Note that χ(·) takes two

individuals as input and returns two modiﬁed individ-

uals. The mutation function ω(·) operates on a single

individual, and it is the composition of two different

levels of mutation. The upper level is that of the or-

dered set of lamp descriptions, and it is mutated as

following:

(I ) =

(

I \ L

if r < 0.5

{

L+1

}

if r > 0.5

(2)

where r, here and in all the following equations, is a

random number in range 0 ···1, i is a random integer

in range 1···L. The lamp description L

L+1

is a new

lamp taken randomly from the set of possible lamps.

Mutation at the lower level, that of single lamp de-

scription, is given by:

(L) =











{

C,W

}

, p, l, w, k

if r > π

{

C,W

}

, p + ∆p, l, w, k

if r > π

{

C,W

}

, p, l

, w, k

if r > π

{

C,W

}

, p, l, w, k

if r > π

(3)

where d

is a new lamp code, extracted randomly from

the set of available types of lamps, l

is a new level of

illumination, selected randomly from the set of possi-

ble light intensities for the lamp of type d, similarly

for k

. The displacement ∆p of lamp positioning is

computed in a random direction from the center p,

and with random offset within a neighbourhood, de-

creased in the course of the evolution. The param-

eters π

{d,p,l,k}

are the mutation probabilities for, re-

spectively, lamp code, lamp position, lighting level,

and color temperature.

During the evolution the entire population is re-

placed, P

(t+1)

← P

(t)

, using the crowded tournament

selection strategy described above. The size of the

population remains constant during evolution. In this

last equation the tournament dominance criterion is

based on two contrasting objectives: the quality of

the light, and the saving of energy. In turn, the light

quality is computed as the combination of two objec-

tives: achieving an illuminance level closest as possi-

ble to the given target, an obtaining light distribution

uniform enough in the given space. The evaluation

of light quality is preformed on samples S, surfaces

distributed in the interior space, that can be placed

in ways convenient to ensure best lighting quality in

key portions of the space. Compliance with the target

level of light, and degree of uniformity, are combined

in a single ﬁtness f

of the individual I, with the fol-

lowing computations:

t (I) =

∑

i=0

− T

(4)

u(I) =

∑

i=0



− S



(5)

(I) = wt(I) + (1 − w)u(I) (6)

where S

is the illumination measured on the sample

produced by the lighting conﬁguration of individual

I, and M is the number of samplers S. Note that treat-

ing t(I) and u(i) as separate ﬁtness in multi-objective

optimization would be incorrect, because are not con-

ﬂicting. It can be easily veriﬁed in the limit case of

an individual

I that illuminates all samplers exactly at

target level T , from equations (4) and (5) we obtain

I) = u(

I) = 0. The weight w control the balance be-

tween the desired compliance with the target level of

light and uniformity, the default value used in all re-

ported results is 0.5. Energy consumption represents

the second ﬁtness and it is quantiﬁed as the overall

power consumption of the lamps (measured in Watt)

divided by the volume of the room:

(I) =

∑

i=0

(7)

where C

is the amount of Watts consumed by the i-th

lamp of the individual I, V the volume of the interior

environment in m

, and N the number of lamps com-

posing the solution.

In the presented problem of lighting optimization

there are some conditions on the design process to be

satisﬁed, therefore a constraint handling method has

to be considered as well. The constrains in question

concern positioning the lamps inside the interior en-

vironment, where a lamp should be mounted on the

walls or on the ceiling in accordance with its model of

light ﬁxture and in contact with the room surface, and

two lamps cannot be placed in the same location. Fur-

thermore, depending on the room design, there might

be some areas where the lamp placement is not al-

lowed, for example in presence of windows, pillars,

or supporting beams. The constraint speciﬁcations

are provided to the system within the 3D model of the

environment itself. The walls and ceiling are struc-

tured as a discrete grid of vertices, each representing

a feasible position for a lamp. With this approach,

the set of constraints can be effortlessly reformulated

for different experiments, ensuring absolute ﬂexibil-

ity in the design process. Since the satisfaction of the

above constraints is mandatory for the problem, they

can be referred as hard constraints. To handle them,

we adopted a strategy based on preserving feasibility

of solutions, where crossover and mutation operations

are speciﬁcally designed to always produce feasible

offspring from feasible solutions.

3 BLENDER AS SIMULATION

ENVIRONMENT

For the purpose of evaluating a lighting conﬁguration,

a virtual environment able to accurately reproduce

the architectural space and its spectral reﬂectometric

properties is needed. Moreover, a physical simula-

tion platform must be considered as well for correct

illumination calculation in sample points of the archi-

tectural space.

This paper investigates the adoption of the 3D

graphic software Blender as a uniﬁed solution to the

two requirements stated above. Firstly, Blender is

the most comprehensive open-source 3D computer

graphic tool available. It is particularly suitable for

modeling architectural interiors, with the possibility

of importing components from CAD ﬁles. Secondly,

Blender provides a physically based rendering engine,

named Cycles, able to exhaustively evaluate lighting

conﬁgurations needed for solving the inverse lighting

problem. Moreover, Blender embeds a Python inter-

preter, which can run scripts supplied by the user, in

order to extend its functionalities. Thanks to its intrin-

sic versatility, Blender has already been applied to a

number of different problems, from the medical ﬁeld

(Daenzer et al., 2007) to industrial applications (Plebe

and Grasso, 2016).

The algorithm here presented has been imple-

mented in the form of a Blender script, composed

of 9 main Python modules. The ﬁrst group of mod-

ules, which rely on Blenders modeling features, per-

forms the simulation environment set-up. The archi-

tectural interior scene of interest is represented inside

the computer graphics software by means of geomet-

ric meshes and material shaders. The room structure

(walls, ﬂoors, ceiling) and its furnishings are deﬁned

by the meshes, while colors, textures and reﬂectiv-

ity properties of the objects are speciﬁed through the

shaders. When evaluating the ﬁtness of a solution,

the 3D scene is enriched with further supporting ele-

ments: the proposed lamps illuminating the environ-

ment, and basic 3D structures employed to perform

individual lighting measurements at locations of inter-

est. Using a sophisticated ray-tracing render engine,

Blender executes an accurate simulation of illumina-

tion, taking into account a variety of environmental

factors. The second group of python modules to ex-

tract light intensity values and their distribution across

the interior space processes the obtained rendered im-

ages.

These outputs are used, in the third group of mod-

ules, by the genetic algorithm to compute the actual

ﬁtness values of a solution. After evaluating the en-

tire current population and selecting the mating pool,

the genetic operators of crossover and mutation are

applied to generate the offspring. The operators are

speciﬁcally implemented for the presented case prob-

lem, as mentioned in in the previous sections, with the

support of an evolutionary computation python

Quality

Consumption

Quality

Consumption

Quality

Consumption

Quality

Consumption

Figure 1: Final populations of the optimization in the two case studies: the coffee shop on the left, and the shopping mall on

the right. The plots in the upper row are the complete populations at the end of the optimization, and the Pareto fronts, in

the lower row there are the results of the clustering, solutions with the star mark are the representative solutions of the four

clusters.

framework named DEAP (Fortin et al., 2012), which

allows to freely customize any component of the ge-

netic algorithm workﬂow. At the end of the execu-

tion of the algorithm, the obtained result is the Pareto

front of the ﬁnal population, namely the set of non-

dominated solutions, each one of them representing

an optimal lighting conﬁguration for the given inte-

rior environment. Optionally, a photorealistic render-

ing of the illuminated scene can be generated.

4 CLUSTERING THE FINAL

POPULATION

As in most multi-objective optimization problems,

our lighting design system typically generates too

many solutions in the ﬁnal Pareto set, and selecting a

single one that best reﬂects the preferences of the ar-

chitect can be a daunting task. A considerable amount

of research effort has been devoted to alleviate this in-

convenience in the general multi-objective case, with

several proposed methods that reduce the Pareto op-

timal set to a set of solutions that is attractive to the

decision maker. A large part of the proposed methods

assumes that the preferences of the decision maker are

well known in advance, and can be expressed in math-

ematical terms and incorporated in the optimization

algorithm (Jaimes and Coello, 2013; Bechikh et al.,

2015). The situation of the architectural lighting de-

sign is different. Although the objectives deﬁned in

our optimization problem capture important require-

ments of the design process, there are aesthetic and

stylistic components of the design process that elude

mathematical formulations. The great advantage of a

tool like the one here proposed is for the architect to

drastically restrict the search space of solutions, and

to concentrate his or her creativity on a small number

of simulated solutions. It is difﬁcult to prescribe in

advance any preferred part of the Pareto front, in prin-

ciple the entire front can offer attractive solutions to

the lighting designers, the choice is up to their exper-

tise and aesthetic disposition. For this reason we fo-

cused on methods commonly classiﬁed as a posteriori

(Zio and Bazzo, 2010), where the selection of a small

subset of solutions is made on the entire ﬁnal approx-

imate Pareto front, computed without the incorpora-

tion of preferences from the the decision maker.

First, we partitioned the set of solutions into a

predeﬁned number of clusters N

, using the subtrac-

tive clustering algorithm (Chiu, 1994; Zio and Bazzo,

2012). Let us deﬁne O the set of vectors in the ﬁtness

space of the ﬁnal solutions F : O = {f (S )|S ∈ F }.

The vectors are normalized with all dimensions in

range [0, 1], we call

O the set of normalized vectors.

For each solution a “potential” function ψ is intro-

duced, that captures the neighborhood size of the so-

lutions: ψ

(0)

) =

∑

o∈

−o

, o

∈

O. The su-

perscript (0) is meant because the previous equation

provides the initial values of the potentials, which are

updated recursively, each time identifying as a cluster

center the solution with the largest potential:

= argmax

o∈

(k)

(o)

, (8)

∆ψ

(k)

) = e

−c

(k)

), o

∈

O, (9)

(k+1)

) = ψ

(k)

) − ∆ψ

(k)

), o

∈

O.(10)

Equation (8) computes the center of the k-th cluster,

the recursive loop is terminated when k = N

, the pre-

deﬁned number of clusters. The parameters r

and r

act effectively as radii, inﬂuencing, respectively, the

range of neighborhood of a solution and the close-

ness of distinct cluster. Their values are computed

as a function of the number of desired clusters N

, r

2.5

All solutions S in F are partitioned in the clus-

ters according to the distance of the vectors in ﬁtness

space to the cluster centers. Calling

(k)

the solution

in F that is center of cluster k, corresponding to the

normalized vector c

, the partitioning is done as fol-

lowing:

Q =





S : arg max

k∈[1..N

]



f(S) − f(

(k)

)



= 1



··· ,



S : arg max

k∈[1..N

]



f(S) − f(

(k)

)



= N



(11)

The solutions offered to the architect for her subjec-

tive evaluation and ﬁnal decision are the centers of the

clusters.

5 RESULTS

We evaluated empirically our lighting optimization al-

gorithm on two case studies. As discussed in the In-

troduction, a satisfactory lighting quality is highly de-

pendent on the visual tasks that are to be performed

in the interior space, and on speciﬁc requirements

of visual interest within the space. These speciﬁca-

tions are passed to the model with the placement of

the samplers and ﬁxing the target illumination level.

All genetic parameters of the model have been tuned

in a preliminary phase on simpler and smaller rooms,

and these settings did not required further tweaking in

the two case studies. The chosen case environments

are both complex architectural interiors, with irreg-

ular and non-convex planimetries, demonstrating that

there are no limitations in the ﬂexibility of application

of the presented system. The ﬁrst case study environ-

ment is the interior of a coffee shop. The architec-

ture of this room has size of 14 × 10 × 2.8 meter, and

it is characterized by a long and narrow dining area

leading to a wider space with a lounge room and a

bar counter. A total of 13 samplers have been used

to evaluate illumination levels, placed in key areas

where light should create visual interest. The genetic

algorithm has been run with a population of 200 indi-

viduals, the ﬁnal population is shown in Fig. 1, where

it is possible to appreciate how the solutions smoothly

span a large Pareto front of the two ﬁtness. The ﬁnal

solutions are clustered in order to provide a small and

manageable subset of solutions, we used four clus-

ters in both the cases here experimented. It is then a

designer choice to pick a desired solution among the

four proposed, as a tradeoff between lighting quality

and energy consumption. The Fig. 2 shows photo-

realistic renderings of two solutions belonging to the

Pareto front. The second case study is the hall of a

shopping mall, of size 12 × 11 × 4.0 meter. It is com-

posed of a central area connected to secondary small

shop. The main space contains a column with display

stands and an area serving as lounge room, while the

secondary area for the small shop has a lower ceiling

level and contains several product racks and a counter

with the cash register. A total of 14 samples have been

used, with a genetic population of 200 individuals.

The Fig. 3 shows two optimal solutions of lighting

conﬁgurations. As in the previous case study, there

is a wide and smooth coverage of the Pareto front.

However, as can be seen in Fig. 1, the Pareto front of

this case study reached even better levels of consump-

tion ﬁtness than the previous one. This result can be

explained by the brighter shading of walls and ﬂoors

in the mall environment (pale yellow and white) re-

ﬂecting more light than the deep red and beige color

tones of the coffee shop, which requires more intense

light sources in order to reach the same perceived il-

lumination level. Nonetheless, the visual results are

rather satisfying in both case studies, demonstrating

how the presented algorithm can be a suitable tool to

effectively design light conﬁguration for a variety of

different environments, with minimum effort from the

user.

Figure 2: Two interior views of two different optimal lighting conﬁgurations in the coffee shop environment.

Figure 3: Two interior views of two different optimal lighting conﬁgurations in the shopping mall case study.

6 CONCLUSIONS

Most part of life in industrial contemporary society

is spent indoor, with activities going on even when

daylight is over. Therefore there is a demand for

artiﬁcial lighting, which is often a critical compro-

mise between the achievement of light level allowing

full photopic vision everywhere in the interior space,

and energy consumption. The strategy here proposed

takes as input an arbitrary layout of interior space, in-

cluding realistic furniture and materials, and a list of

possible realistic light sources, generating as output

solutions, optimal under the compliance with the tar-

get illumination level, and the consumption of elec-

tric power. However, for designing an accurate in-

terior illumination and able to not be glare nor inad-

equate, it is needed to optimize multiple constraints,

often conﬂicting, which make hard the use of classical

computational methods. Thus, a multi-objective ge-

netic algorithm has been developed for interior light-

ing design, with the main aim to (1) optimize the level

of luminous intensity, and (2) maximizing the energy

saving. Moreover, a clustering of solutions approach

has been also developed, in order to reduce the search

space and the Pareto front, helping, then, the deci-

sion maker in the choice and selection of the more

appropriate illumination. In combination to NSGA-

II, the proposed computational tool is based also on

a 3D graphic software, that is Blender, for providing

a rendering engine for direct illumination and repro-

duce the architectural space to be lighted. Two dif-

ferent case studies have been considered in order to

evaluate the accuracy, and efﬁciency of the illumina-

tion produced, based on different complex shapes of

the architectural interiors (irregular and non-convex),

which make harder the design of an uniform illumina-

tion distribution. Finally, from the analysis of the sev-

eral experiments performed, the presented algorithm

has showed to be a suitable and effective tool for in-

terior lighting design in a variety of different environ-

ments.

REFERENCES

Bechikh, S., Kessentini, M., Said, L. B., and Gh

edira, K.

(2015). Preference incorporation in evolutionary mul-

tiobjective optimization: A survey of the state-of-the-

art. Advances in Computers, 98:141–207.

Bertoldi, P., Hirl, B., and Labanca, N. (2012). Energy efﬁ-

ciency status report. Technical report, European Com-

mission – Institute for Energy and Transport.

Caldas, L. (2016). Painting with light: An interactive evo-

lutionary system for daylighting design. Building and

Environment. doi: 10.1016/j.buildenv.2016.07.023.

Chiu, S. L. (1994). Fuzzy model identiﬁcation based on

cluster estimation. Journal of Intelligent & fuzzy sys-

tems, 2:267–278.

Cuttle, C. (2015). Lighting design: a perception-based ap-

proach. Routledge, New York.

Daenzer, S., Montgomery, K., Dillmann, R., and Un-

terhinninghofen, R. (2007). Real-time smoke and

bleeding simulation in virtual surgery. In Westwood,

J. D., Haluck, R. S., Hoffman, H. M., Mogel, G. T.,

Phillips, R., Robb, R. A., and Vosburgh, K. G., ed-

itors, Medicine Meets Virtual Reality, pages 94–99.

IOS Press, Amsterdam.

Deb, K., Agrawal, S., Pratap, A., and Meyarivan, T. (2000).

A fast elitist non-dominated sorting genetic algorithm

for multi-objective optimization: NSGA-II. In In-

ternational Conference on Parallel Problem Solving

From Nature, pages 849–858.

Fortin, F.-A., De Rainville, F.-M., Gardner, M.-A., Parizeau,

M., and Gagn

e, C. (2012). DEAP: Evolutionary algo-

rithms made easy. Journal of Machine Learning Re-

search, 13:2171–2175.

Gordon, G. (2014). Interior lighting for designers. John

Wiley, New York.

Jaimes, A. L. and Coello, C. A. C. (2013). Interactive ap-

proaches applied to multiobjective evolutionary algo-

rithms. In Doumpos, M. and Grigoroudis, E., editors,

Multicriteria decision aid and artiﬁcial intelligence:

Theory and applications, pages 191–207. John Wiley,

New York.

Livingston, J. (2015). Designing with light: the art, science,

and practice of architectural lighting design. John Wi-

ley, New York.

Palmer, S. (1999). Vision Science – Photons to Phe-

nomenology. MIT Press, Cambridge (MA).

Plebe, A. and Grasso, G. (2016). Simulating ﬁre outbreaks

in industrial plants. In Blender Conference, Amster-

dam.

Sansoni, P., Farini, A., and Mercatelli, L., editors (2015).

Sustainable Indoor Lighting. Springer-Verlag, Berlin.

Wunderlich, C.-H. (2003). Light and economy: an essay

about the economy of prehistoric and ancient lamps.

In Chranovski, L., editor, Lychnological news, pages

251–264. LychnoServices, Hauterive (Suisse).

Zio, E. and Bazzo, R. (2010). Multiobjective optimization

of the inspection intervals of a nuclear safety system:

A clustering-based framework for reducing the pareto

front. Annals of Nuclear Energy, 37:798–812.

Zio, E. and Bazzo, R. (2012). A comparison of methods

for selecting preferred solutions in multiobjective de-

cision making. In Kahraman, C., editor, Computa-

tional Intelligence Systems in Industrial Engineering

With Recent Theory and Applications, pages 23–43.

Atlantis Press, Paris.