Energy and Buildings 92 (2015) 195–203

Contents lists available at ScienceDirect

Energy and Buildings

journal homepage: www.elsevier.com/locate/enbuild

Evolutionary-driven search for solar building models using LiDAR data

Marko Bizjak ∗ , Borut Žalik, Niko Lukač
University of Maribor, Faculty of Electrical Engineering and Computer Science, Smetanova ulica 17, SI-2000 Maribor, Slovenia

a r t i c l e i n f o a b s t r a c t

Article history: The search for solar buildings is one of the primary challenges in urban planning, especially when devel-
Received 28 October 2014 oping self-sustainable cities. This work uses an evolutionary approach for finding the optimal building
Received in revised form 21 January 2015 model based on airborne Light Detection And Ranging (LiDAR) laser-scanned data, regarding solar poten-
Accepted 24 January 2015
tial. The method considers self-adaptive differential evolution for solving the constrained optimisation
Available online 2 February 2015
problem. In the experiments, the effect of different buildings’ layouts and design parameters were ana-
lysed regarding solar irradiance. Rectangular, T and L-shaped buildings were considered with various
Keywords:
design parameters: position, building rotation, facades’ height, roof’s height and slope. The experiments
Solar energy
Solar building
confirmed that the method can efficiently find the solar building design with maximum solar potential
Differential evolution within constrained optimisation space.
LiDAR data © 2015 Elsevier B.V. All rights reserved.

1. Introduction
Solar energy is one of the most available renewable and clean
types of energy [1]. Nonetheless, its great potential for electrical
energy generation remains widely unutilised. In order to develop
low energy buildings for self-sustainable cities, reduce carbon
emissions or maximise passive solar heating, it is vital to know
how to maximise solar energy utilisation. Solar energy is normally
actively converted into electricity using photovoltaic (PV) systems.
Since the recent development of cheaper and more efficient PV
modules [2,3], these systems have received significant positive
attention. They are commonly installed on buildings’ roofs, where
it is generally considered that a surface oriented towards the Equa-
tor with a tilt angle equal to a location’s latitude is optimal for solar
energy utilisation [4]. However, this is often not the case due to
local climatic conditions [5] and the influence of shadowing from
terrain and man-made objects that decrease the received irradi-
ance. Therefore, the optimal slope and orientation of a PV system
attached onto a building’s surface presents an optimisation issue for
investors, as well as for architects, urban planners and civil engi-
neers who want to assess the solar energy potential of rooftops,
renewable energy integration into buildings and the carbon foot-
print of buildings.
Over recent years several methods have been developed in
order to address this issue by modelling solar irradiance on a
given surface. This is an increasing problem for energy-efficient
∗ Corresponding author. Tel.: +386 2 220 7435; fax: +386 2 220 7272.
E-mail address: m.bizjak@um.si (M. Bizjak).
building design and for finding the most suitable building’s sur-
faces for PV systems’ placements. Some studies [6–8] only consider
self-shadowing and not the shadowing from surroundings. This is
evident when considering the modelling of buildings within urban
areas, where solar radiation obstructions are more common [9].
With the advancement of remote sensing technologies, such as
Light Detection And Ranging (LiDAR), new methods have been
developed for buildings’ irradiance modelling and for estimating
the solar potential of buildings’ roofs [10]. LiDAR is an active remote
sensing technology that can be mounted on an aircraft in order to
scan the Earth’s surface using laser pulses. The resulting LiDAR data
is presented as an unstructured set of 3D points (i.e. point cloud).
Various studies have been conducted in the search for the most
efficient solar building design. Most methods [11,6–8,12–14] con-
sider a selected range of parameters (e.g. predetermined roofs’
slopes). Others [15–18] have considered automatic approaches that
are presented as optimisation problems. This is generally solved
by evolutionary-driven algorithms. Hachem et al. [7] performed a
parametric investigation of building design regarding solar poten-
tial. In another study [6], the authors developed a methodology of
neighbourhood design that included streets’ and different build-
ings’ shapes, and their positions in order to avoid mutual shading.
Furthermore, they used this methodology for evaluating the energy
supply for neighbourhoods housing [8]. Energy demand was com-
pared with the solar potential of different neighbourhoods by
considering heating and cooling requirements. Kämpf et al. [16]
presented a new evolutionary algorithm for optimising solar irra-
diation availability. Its performance was compared with several
related evolutionary algorithms, including DE. They searched for
optimal positions for multiple buildings but did not consider their

http://dx.doi.org/10.1016/j.enbuild.2015.01.051
0378-7788/© 2015 Elsevier B.V. All rights reserved.
196 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203
Nomenclature
x population vector
f fitness function
Hi upper bound
Ic solar potential for a cell at a given time
Ig global solar irradiance
Icb terrestrial direct irradiance of a given cell
Icd terrestrial diffuse irradiance of a given cell
Jc daily solar insolation of a given cell
Li lower bound
Mi i-th location at mountainous LiDAR dataset
Sc shadowing coefficient
Ui i-th location at urban LiDAR dataset
xi i-th element of the population vector x
Abbreviations
CR crossover rate
DE differential evolution
F amplification factor of the difference vector
G generation
IDW inverse distance weighting
LiDAR light detection and ranging
P population size
PV photovoltaic
SPA solar positional algorithm
design. Moreover, Kämpf et al. [17] presented a new methodology
for optimising building and urban geometric forms for the utilisa-
tion of solar irradiation. Kämpf et al. [18] proposed an alternative
evolutionary algorithm, a multi-objective optimiser, in order to
optimise several parameters, such as building’s height, roof’s height
and its orientation. The main objective was to maximise avail-
able solar irradiation for new urban forms. Ouarghi and Krarti [15]
presented the optimisation of an office building’s shape regarding
energy and construction costs using a genetic algorithm and an
artificial neural network. Esch et al. [12] studied the influence of
street and building design parameters such as street width, orien-
tation and building’s envelope design regarding the received solar
energy. Ling et al. [11] investigated the effect of the geometric shape
and orientation of high-rise buildings in order to minimise solar
irradiation. Hwang et al. [13] examined the maximisation of solar
irradiation utilisation for high-rise buildings with installed PV sys-
tems. In their study they analysed the effects of the inclination and
direction of PV systems, as well as the distance to the module length
ratio. Roohollahi et al. [19] inspected solar energy gain on seven 3D
geometries during different months of the year. Kanters and Wall
[14] studied the effect of various building blocks’ design including
urban density on the received solar radiation. To our knowledge
none of these methods have considered LiDAR data in order to
estimate solar efficient buildings.
Several methods have been developed for solar radiation mod-
elling [20] that mainly differentiate on the estimation of the diffuse
irradiance. During recent years, a number of methods for solar
potential estimation have been developed that consider topog-
raphy from LiDAR data [21–29,10,30–32]. With the advancement
of remote sensing technologies and evolutionary computing, new
methods can be developed for solving the irradiance optimisation
problem regarding building design, by considering real data.
This paper presents a novel optimisation of a building model
using LiDAR data, in order to maximise a building’s received
solar irradiance by finding the most solar-efficient design. The
usage of LiDAR data allows us to perform the search for opti-
mal solar building within a real urban environment, where several
optimisation constraints (e.g. shadowing from buildings and ter-
rain, local climate, and terrain topography) are considered. The real
environment, provided by LiDAR data, presents a difficult problem
when searching for global optimum regarding solar potential, as
there are many factors that affect it. Moreover, the optimisation
of a building’s model is a high-dimensional problem with multiple
parameters to be considered, where manual inspection would be
exhausting. Although many other optimisation algorithms could be
used, DE (differential evolution) [33] was selected, as it is an evo-
lutionary method for finding global optimum within a reasonable
time even when the problem space is large and contains many local
extremes. The presented method consists of two core stages. Dur-
ing the first stage the user provides a specific building model that
is integrated into existing LiDAR data. During the second stage the
optimisation problem is solved by using self-adaptive DE [34]. This
method considers the following optimisation parameters for build-
ing modelling: orientation, height, location and roof’s height and
slope. The maximisation criterion is the estimated solar irradiance
[10] based on shadowing from surrounding objects, terrain and
vegetation, as well the local climate conditions that are captured by
the long-term diffuse and global on-site irradiance measurements
using a pyranometer.
The paper is structured into 4 sections. The next section
describes the proposed method for optimising solar building
design. The results of the experiments are presented in Section 3.
The last section concludes this paper.
2. Method for modelling solar-efficient buildings
The proposed method is described in detail in the following
subsections. Section 2.1 presents parameter-based building mod-
elling, whilst Section 2.2 describes the used evolutionary approach.
The approach is based on the self-adaptive DE with heuristics that
speed-up the search for the most solar efficient building model
within the constrained solution space.
2.1. Building modelling within LiDAR data
The method’s input is the georeferenced LiDAR point cloud,
where each point is classified as either building, terrain or vegeta-
tion [35–37], as shown in Fig. 1a. Firstly, the point cloud is arranged
into a regular 2.5D grid. In order to improve the accuracy of the solar
potential estimation for each building, the empty cells (i.e. without
points) are interpolated using the inverse distance weighting (IDW)
method [38]. The height of each cell is defined by the highest point
of the point cloud’s subset located within the given cell. Fig. 1b
shows the 2.5D grid together with a polygonal building model.
A user-defined building is interactively modelled in two steps.
Firstly, the user imports a 3D building model, which is then con-
verted into virtual 2.5D blocks that define the building’s shape from
the top-down perspective. The width and length of a block are
defined as a multiple of the grid’s resolution. The building’s shape
from this perspective remains constant throughout the optimisa-
tion process. In the next step all blocks are normalised and placed
at the same height as the lowest block in order to embed the build-
ing onto the surface. The rest of the building’s design is defined by
several key parameters: position, building rotation, facades’ height,
roof’s height (i.e. vertical distance from walls to the ridge) and slope,
as shown in Fig. 1b. The building’s position is within the area of
interest for the building’s location. The area of interest is defined
as a polygonal area on the grid selected by the user. The build-
ing’s rotation is considered around the centre of its bounding box.
The roof’s height allows alternation of the rooftop design from the
ridged to a flat horizontal surface at a given roof height. The roof’s
 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203 197

Fig. 1. (a) Classified LiDAR data; (b) Building model on the 2.5D grid of classified LiDAR data, where the following building design parameters are considered: r – rotation
axis, f – facades’ height, h – roof’s height and ˛ – roof’s slope; (c) Building’s cells within the 2.5D grid.

slope is measured from the upper surface of a block, where a given
roof can be inclined from all sides of each block.
After the building has been modelled, it is rasterised into the
2.5D grid. During the rasterisation process, for each covered cell,
the highest point of the roof is considered within the given cell. The
covered cells represent the building within the grid (see Fig. 1c).
2.2. Proposed evolutionary approach
crossover, the parameter values are taken from the first parent as
long as the random number is smaller than CR. Otherwise, all of
the remaining parameters are obtained from the second parent
[39].
In this work, self-adaptation of F and CR control technique, as
proposed by Brest et al. [34], is used in order to achieve more flexible

The main challenge is the global optimisation of parameters that

describe the modelled building in order to maximise the received
solar irradiance. The optimisation problems are generally solved
using numerical solutions of evolutionary algorithms, which are
inspired by biological evolution [34]. One of the known approaches
is DE as proposed by Price and Storn [33]. DE is robust against the
convergence in local minimums due to random mutation. Further-
more, it is suitable for problems where it is difficult to find a solution
due to the large solution space, such as the case considered in this
paper.
The global optimisation is generally defined as the search for
the supremum of f(x ), where f is the fitness function and x ∈ Rn ,
x =[x1 ,x2 ,. . .,xn ] is the solution vector, where each element is limited
by the lower and upper bound xi ∈ [Li ,Hi ] [39]. In terms of evolu-
tionary optimisation, x is called a candidate or a population vector
and xi an optimisation parameter. DE can be considered as a direct
parallel search method [33] that uses P n-dimensional vectors
xi,G ; i = 1, . . ., P as a population that evolves with each generation
(i.e. iteration) G.
New candidates are created by merging a candidate’s parent and
some other vectors in the same population. Candidates are selected
based on the fitness function which, in the case of this paper, is a
solar potential estimation as summarised in the next subsection. DE
is controlled using three control parameters: amplification factor
of the difference vector F, crossover rate CR and population size P.
CR can be described as a probability for a parameter to be obtained
from the parent or in other words CR controls the parent’s influence
on the offspring [39].
Several DE variants exist and are generally classified using the
notation DE/a/b/c, where:

• a determines the vector to be mutated, which could be rand (ran-

domly selected vector) or best (vector with the best fitness);
• b stands for the number of difference vectors;
• c defines the crossover operation [33]. In this paper, binomial
(bin) and exponential (exp) types of crossover are considered. The
main difference between them is that for the binomial type every
parameter value is taken from one of the parents determined
by the crossover rate (CR). On the contrary, in the exponential Fig. 2. Workflow of the proposed method for the optimisation of a building model.
198 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203

Fig. 3. Example of (a) classified Lidar data in 2.5D grid and the per-cell (b) shadowing coefficient Sc , (c) direct irradiance Icb , (d) diffuse irradiance Icd and (e) solar potential Ic .

optimisation. F and CR parameters are adapted for each vector as During the mutation, for each target vector xi,G , i = 1, . . ., P, a
follows: mutated vector is generated by adding the weighted difference
 between two vectors to the third vector of the population:
Ft + rand1 · Fu , if rand2 < 1
Fi,G+1 = , (1) vi,G+1 = xr1 ,G + F(xr2 ,G − xr3 ,G ), (3)
Fi,G , otherwise
 where r1 , r2 , r3 ∈ {1, 2, . . ., P} are random integers; r1 = / r2 =
/ r3
rand3 , if rand4 < 2 ⇒ P ≥ 4 and F ∈ [0,2] represents an amplification of the difference
CRi,G+1 = , (2) vector (xr2 ,G − xr3 ,G ) [33]. The crossover operation is performed fol-
CRi,G , otherwise
lowing the mutation operation.
where randj ∈ [0,1], j ∈ {1, 2, 3, 4}.  1 and  2 represent the proba- The purpose of the crossover operation is to increase the vec-
bility adjustment factors for F and CR, respectively. In this work the tor (i.e. candidate) diversity. In order to obtain the trial vector, the
following values are used:  1 =  2 = 0.1, Ft = 0.1 and Fu = 0.9, as sug- target vector is crossed with the mutated vector:
gested in [34]. The most common strategy is DE/rand/1/bin [33]. ui,G+1 = (u1i,G+1 , u2i,G+1 , . . ., uni,G+1 ), (4)
Three operations are performed for each vector in every gener- 
ation: mutation, crossover, and selection. The mutation and the vji,G+1 , if (randj ≤ CR) or j = rni
crossover are used for generating new candidates, whilst selection uji,G+1 = , (5)
xji,G , if (randj > CR) or j =
/ rni
determines which vector shall survive in the current generation.

Fig. 4. LiDAR dataset representing an urban area with designated locations U1 and U2.
 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203
Fig. 5. LiDAR dataset representing a mountainous area with designated locations M1 and M2. A part of M2 location is concealed by the hill.
where randj is the output of a random generator, CR ∈ [0,1], rni is
a randomly-selected index for ensuring that ui,G+1 obtains at least
one parameter from the mutated vector vi,G+1 [33]. Otherwise, a
new population vector would not be created and the population
would cease to evolve [34]. If a solution parameter of a vector is
outside the parameter limits then, as a compromise, the parameter
value is set at the mean value of the parameter limits. During the
selection operation, the trial vector ui,G+1 replaces the target vector
xi,G in case it yields a higher solar potential estimation [33]:
 required in order to estimate the slope and aspect angles.
ui,G+1 , if f (ui,G+1 ) > f (xi,G )
ui,G+1 = . (6)
xi,G , otherwise
In this operation, the building is modelled based on the parameters
of a given candidate. For each candidate, the original grid remains
intact, whilst a different building design is considered on the 2.5D
grid. The cells covered by the building are then used for solar poten-
tial estimation. Furthermore, a problem of non-fixed number of
covered cells is encountered due to rasterisation after the rotation.
The minimisation of the number of covered cells is done by the
adjustment of the building’s position and additional elimination of
the excess cells. The original number of covered cells is set when
the building is not rotated. The excess cells are detected after the
rotation and iteratively removed from the building’s edges oriented
towards the Equator starting from the right side, until the number
of cells is matched to the original amount. The presented selection
is performed in the same way for every tested strategy. The criteria
(i.e. fitness function) used in the selection is described in the next
subsection.
2.2.1. Maximisation of the solar irradiance
The initial population is randomly generated over the entire
parameter space [33] for all parameters apart from building rota-
tion and roof slope. These two have proven to have significant
impact on solar potential estimation and can be approximated to
some extent, which in turn speeds-up the search. Generally, a sur-
face oriented towards the Equator with a tilt angle equal to the
location’s latitude is optimal for solar energy utilisation [4]. Hence, a
simple heuristic is proposed where the initial population parameter
space for roof slope is considered in Gaussian distribution, where
the peak of the distribution equals the location’s latitude. A similar
heuristic is used for the building’s rotation parameter, where the
distribution’s peak is oriented towards the Equator. The workflow
of the presented method is shown in Fig. 2.
199
The considered fitness function runs the method for solar poten-
tial estimation [10], which considers 2.5D grid as the input, based
on the underlying LiDAR point cloud. Additional input are the cells
of a candidate building. In order to achieve greater performance, a
parallel implementation of the method on graphics processing unit
(GPU) [40] was used. The solar potential calculation is performed
in the following steps:
1. Normals for each building’s cell are calculated. These are
2. The per-cell time and location dependent terrestrial irradiance
calculation is performed by considering the cell’s aspect and
slope angles [4], as well as the long-term on-site diffuse Id and
global Ig solar irradiance measurements by the pyranometer. The
direct irradiance is simply calculated as Ig − Id .
3. The final estimation of the solar potential for a cell at a given
time is then defined as:
 kWh 
Ic = Icb (1 − Sc ) + Icd , (7)
m2
where Icb and Icd are the terrestrial direct and diffuse irradi-
ances of a given cell, whilst Sc is the shadowing coefficient that
decreases the direct irradiance. Fig. 3 shows each influence factor
considered for solar potential estimation.
The Sun’s position is required for accurate shadowing, and is
obtained with the precise Solar Positional Algorithm (SPA) [41].
4. Time-dependant irradiance Ic is considered between sunrise Tsr
and sunset Tss with the fixed time-step, in order to estimate the
daily solar insolation:
 Tss  kWh 
Jc = Ic (t)dt . (8)
m2
Tsr
5. The solar potential is defined as an average daily insolation
throughout the year [10].
6. The final result of the fitness function is the sum of the solar
potential for each cell of the building. Hence, this presents the
average amount of the daily solar energy a building receives in
kWh/m2 .
3. Results and discussion
The presented method was tested on two different LiDAR
datasets, one representing a generally flat urban environ-
ment (46◦ 6 21 N, 14◦ 31 53 E), whilst the other a mountainous
200 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203

Rectangular T-shape L-shape

0.668
0.66
Solar potential [M W/h]

0.524

Solar potential [M W/h]

Solar potential [M W/h]

0.658
0.666
0.522
0.656
U1

0.520
0.654 0.664

0.518 0.652

0.662
0 3 6 9 12 15 18 21 0 10 20 30 0 10 20 30
Generation Generation Generation

0.529 0.674
0.658
Solar potential [M W/h]

Solar potential [M W/h]

Solar potential [M W/h]

0.672
0.528 0.657
0.67
0.527 0.656
U2

0.668
0.655
0.526
0.654 0.666
0.525
0.653 0.664

0 4 8 12 0 11 22 33 44 55 0 7 14 21 28
Generation Generation Generation

0.513 0.628
0.61
Solar potential [M W/h]
Solar potential [M W/h]

Solar potential [M W/h]

0.512 0.626
0.608
0.511 0.624
M1

0.51 0.606 0.622

0.509 0.604 0.62

0.508 0.618
0.602
0 3 6 9 0 11 22 33 44 55 66 77 0 6 12 18
Generation Generation Generation

0.516 0.632
0.615
Solar potential [M W/h]

Solar potential [M W/h]

Solar potential [M W/h]

0.515 0.631
0.614
0.514 0.63
0.613 0.629
0.513
M2

0.628
0.512 0.612
0.627
0.511 0.611 0.626
0.51 0.625
0.61
0 3 6 9 0 7 14 0 2 4 6 8
Generation Generation Generation

rand/1/bin best/1/bin rand/1/exp

Generation best/1/exp best/2/bin

Fig. 6. Best solar potential estimation in each generation for all tested combinations of building shapes and areas of interest using different DE strategies. Solar potential is
estimated as the daily average irradiance of the entire building.

landscape (46◦ 32 30 N, 15◦ 34 6 E), as shown in Figs. 4 and 5, Average measurements from the nearest meteorological station
respectively. Both datasets have northern continental climate. For over the last decade were used in order to calculate the solar poten-
proof-of-concept, the testing was performed using three types of tial for each building candidate. As the self-adaptation technique
building shapes: rectangular, L and T shape. For the L-shape model was used, the population size remained the only parameter to con-
the roof was inclined by the same angle from all sides of each block. trol DE, besides the candidate parameter bounds. The population
During the experiments, every building was tested at four loca- size was set as a multiple of the number of building parameters
tions (i.e. regions U1, U2, M1, M2), two per dataset as shown in by factor 10 (60), as suggested in [33]. As it turned out, this size
Figs. 4 and 5. was insufficient for strategies with the approach best (DE/best/b/c).
 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203 201

300 lines, each representing a single search run using the considered
strategy with the best calculated solar potential per generation.
250
The goal of each search was to quickly find the optimum, which
200 Rect means that searches ending in earlier generations were more suc-
cessful. DE strategies with the approach best (DE/best/b/c) provided
L
150 better performance than others, because the graphs representing
T
them are generally steeper and converge towards the optimum
100 Sum
faster. In most cases they had already gained a noticeable advantage
50
during the first generation. The amount of required generations
to find the optimum differed considerably for each location. It is
0 a consequence of topographical diversity. For example, buildings
rand/1/bin best/1/bin rand/1/exp best/1/exp best/2/bin
at location M2 were found faster than elsewhere. This occurred
Fig. 7. Sum of the required generations until convergence for each considered DE because they were located on the top of the hill, which means there
strategy and building type. was no shadowing from surroundings. In that case the parameter
facades’ height became irrelevant. Higher buildings are less likely
to be shadowed and consequently have a higher solar potential.
It is suspected that the main reason for this is in the high amount For an easier direct comparison, graphs from Fig. 6 are presented
of local maximums, where shadowing is lower and, consequently, in Fig. 7 with the sum of generations until convergence, based on
DE can converge too quickly towards a local maximum that is the considered strategy and building type.
not necessarily the least shadowed. In order to solve this prob- In Fig. 7 it can be seen that the overall best performer (i.e.
lem, the population size was increased to 80 candidates, which the lowest number of generations required for the search) was
proved to be enough. The search was completed when the differ- the DE/best/1/bin strategy, followed by DE/best/2/bin. It is evident
ence between the calculated solar potential and the global optimum that the optimal rectangular building was found faster than other
was at most 0.25%. In this case, DE satisfied the convergence con- shapes, which is due to smaller number of building cells. On the
ditions. The candidate maximum roof angle parameter was set at other hand, the T and L-shape buildings had the same amount of
70◦ where the roof was still ridged. The rotation was also limited covered cells. However, the L-shape was found quicker and its solar
to 90◦ . potential estimation was slightly higher. This occurred because the
Fig. 6 presents the results for each building type for all consid- internal shadowing between building’s cells was higher for the T-
ered strategies and areas of interest. Every graph consists of five shape building. Internal shadowing takes place when a building’s

Fig. 8. Optimal building models rasterized into the 2.5D grid.

202 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203
cell is shadowed by another cell of the same building. Additionally,
the roof design was different, as the roof was sloped from all sides
of each block for the L-shape building. It can also be observed that
strategies with the approach rand (DE/rand/b/c) benefited from the
exponential recombination (DE/a/exp/c), whereas others did not.
Fig. 8 shows the optimal building models found during testing,
presented in Figs. 6 and 7, for each building type and area of inter-
est. All buildings, tested at a particular area of interest, were located
at approximately the same part of the area. They were generally
oriented on an east-west axis, where the most exposed roof sur-
faces are oriented towards the Equator. The rectangular buildings
were in all cases strictly on the east-west axis (i.e. not rotated),
whereas the T and L-shaped buildings were often rotated by up to
10◦ from the axis. This was mostly due to the internal shadowing of
the building’s cells. All optimal buildings, except those located on
the top of the hill (M2), had the maximum allowed height in order to
decrease the shadowing from surroundings. On the urban dataset,
the building’s roofs were sloped on average at 50◦ (±3◦ ), whereas
at the mountainous location the average slope was 45◦ (±2◦ ). The
difference occurred because surface topography and geographic
location differed for each area of interest. Moreover, on the urban
dataset, where slopes were higher, tested buildings were heavily
shadowed by nearby buildings.
4. Conclusion
This paper proposed a novel method for searching optimal solar
building models within urban areas. For this the differential evo-
lution (DE) was used and tested using various strategies. For every
candidate, a building was modelled and evaluated regarding solar
potential by considering shadowing from real topographic data and
local climate conditions that were captured using pyranometer. To
our knowledge, this is the first attempt to use LiDAR data in order to
find the most efficient building design regarding solar potential. The
results suggest that strategies with the approach best (DE/best/b/c)
are more suitable for this kind of optimisation problem, provided
that the population size is large enough. The best performer turned
out to be the DE/best/1/bin strategy. The presented method could
also be easily extended to handle more building design parameters.
Acknowledgements
This work was supported by Slovenian Research Agency
under grants P2-0041, 1000-13-0552, and J2-6764. Thanks to the
Slovenian Environment Agency for providing the pyranometer
measurements.
Appendix A. Supplementary data
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/j.enbuild.
2015.01.051.
References
[1] M. Dresselhaus, I. Thomas, Alternative energy technologies, Nature 414 (6861)
(2001) 332–337.
[2] A. Luque, S. Hegedus, Handbook of Photovoltaic Science and Engineering, Wiley,
Chichester, West Sussex, England, 2003.
[3] G. Singh, Solar power generation by PV (photovoltaic) technology: a review,
Energy 53C (2013) 1–13, http://dx.doi.org/10.1016/j.energy.2013.02.057.
[4] J.A. Duffie, W.A. Beckman, Solar Engineering of Thermal Processes, Wiley-
Interscience, New York, 2006.
[5] J.D. Mondol, Y.G. Yohanis, B. Norton, The impact of array inclination and ori-
entation on the performance of a grid-connected photovoltaic system, Renew.
Energy 32 (2007) 118–140, http://dx.doi.org/10.1016/j.renene.2006.05.006.
[6] C. Hachem, A. Athienitis, P. Fazio, Investigation of solar potential of housing
units in different neighborhood designs, Energy Build. 43 (2011) 2262–2273,
http://dx.doi.org/10.1016/j.enbuild.2011.05.008.
[7] C. Hachem, A. Athienitis, P. Fazio, Parametric investigation of geometric
form effects on solar potential of housing units, Solar Energy 85 (9) (2011)
1864–1877, http://dx.doi.org/10.1016/j.solener.2011.04.027.
[8] C. Hachem, A. Athienitis, P. Fazio, Evaluation of energy supply and demand
in solar neighborhood, Energy Build. 49 (2012) 335–347, http://dx.doi.org/
10.1016/j.enbuild.2012.02.021.
[9] D. Robinson, A. Stone, Solar radiation modelling in the urban context,
Solar Energy 77 (3) (2004) 295–309, http://dx.doi.org/10.1016/j.solener.
2004.05.010.
[10] N. Lukač, D. Žlaus, S. Seme, B. Žalik, G. Štumberger, Rating of roofs sur-
faces regarding their solar potential and suitability for PV systems based
on LiDAR data, Appl. Energy 102 (2013) 803–812, http://dx.doi.org/10.1016/
j.apenergy.2012.08.042.
[11] C. Ling, M.H. Ahmad, D. Ossen, The effect of geometric shape and building ori-
entation on minimising solar insolation on high-rise buildings in hot humid
climate, J. Constr. Dev. Countries 12 (1) (2007) 27–38.
[12] M. van Esch, R. Looman, G. de Bruin-Hordijk, The effects of urban and build-
ing design parameters on solar access to the urban canyon and the potential
for direct passive solar heating strategies, Energy Build. 47 (2012) 189–200,
http://dx.doi.org/10.1016/j.enbuild.2011.11.042.
[13] T. Hwang, S. Kang, J.T. Kim, Optimization of the building integrated photo-
voltaic system in office buildings’ focus on the orientation, inclined angle
and installed area, Energy Build. 46 (2012) 92–104, http://dx.doi.org/10.1016/
j.enbuild.2011.10.041.
[14] J. Kanters, M. Wall, The impact of urban design decisions on net zero energy
solar buildings in Sweden, Urban Plann. Trans. Res. 2 (1) (2014) 312–332,
http://dx.doi.org/10.1080/21650020.2014.939297.
[15] R. Ouarghi, M. Krarti, Building shape optimization using neural network and
genetic algorithm approach, ASHRAE Trans. 112 (1) (2006) 484–491.
[16] J.H. Kämpf, D. Robinson, A hybrid CMA-ES and HDE optimisation algorithm with
application to solar energy potential, Appl. Soft Comput. 9 (2) (2009) 738–745,
http://dx.doi.org/10.1016/j.asoc.2008.09.009.
[17] J.H. Kämpf, D. Robinson, Optimisation of building form for solar energy utili-
sation using constrained evolutionary algorithms, Energy Build. 42 (6) (2010)
807–814, http://dx.doi.org/10.1016/j.enbuild.2009.11.019.
[18] J.H. Kämpf, M. Montavon, J. Bunyesc, R. Bolliger, D. Robinson, Optimisation of
buildings’ solar irradiation availability, Solar Energy 84 (4) (2010) 596–603,
http://dx.doi.org/10.1016/j.solener.2009.07.013.
[19] E. Roohollahi, M. Mehrabian, M. Abdolzadeh, Prediction of solar energy gain on
3-D geometries, Energy Build. 62 (2013) 315–322, http://dx.doi.org/10.1016/
j.enbuild.2013.03.008.
[20] C. Demain, M. Journée, C. Bertrand, Evaluation of different models to esti-
mate the global solar radiation on inclined surfaces, Renew. Energy 50 (2013)
710–721, http://dx.doi.org/10.1016/j.renene.2012.07.031.
[21] B. Yu, H. Liu, J. Wu, W. Lin, Investigating impacts of urban morphology on spatio-
temporal variations of solar radiation with airborne LiDAR data and a solar flux
model: a case study of downtown Houston, Int. J. Remote Sens. 30 (17) (2009)
4359–4385, http://dx.doi.org/10.1080/01431160802555846.
[22] A. Jochem, B. Höfle, M. Rutzinger, N. Pfeifer, Automatic roof plane detec-
tion and analysis in airborne LiDAR, Sensors 9 (7) (2009) 5241–5262,
http://dx.doi.org/10.3390/s90705241.
[23] J. Hofierka, J. Kanuk, Assessment of photovoltaic potential in urban areas using
open-source solar radiation tools, Renew. Energy 34 (10) (2009) 2206–2214,
http://dx.doi.org/10.1016/j.renene.2009.02.021.
[24] R. Levinson, H. Akbari, M. Pomerantz, S. Gupta, Solar access of residential
rooftops in four California cities, Solar Energy 83 (12) (2009) 2120–2135,
http://dx.doi.org/10.1016/j.solener.2009.07.016.
[25] L. Wiginton, H. Nguyen, J. Pearce, Quantifying rooftop solar photovoltaic
potential for regional renewable energy policy, Computers, Environ. Urban
Syst. 34 (4) (2010) 345–357, http://dx.doi.org/10.1016/j.compenvurbsys.
2010.01.001.
[26] T.R. Tooke, N.C. Coops, J.A. Voogt, M.J. Meitner, Tree structure influences on
rooftop-received solar radiation, Landscape Urban Plan. 102 (2) (2011) 73–81,
http://dx.doi.org/10.1016/j.landurbplan.2011.03.011.
[27] H.T. Nguyen, J.M. Pearce, Incorporating shading losses in solar photovoltaic
potential assessment at the municipal scale, Solar Energy 86 (5) (2012)
1245–1260, http://dx.doi.org/10.1016/j.solener.2012.01.017.
[28] M. Brito, N. Gomes, T. Santos, J.A. Tenedório, Photovoltaic potential in a Lisbon
suburb using LiDAR data, Solar Energy 86 (1) (2012) 283–288, http://dx.doi.org/
10.1016/j.solener.2011.09.031.
[29] J. Hofierka, M. Zlocha, A new 3-D solar radiation model for 3-D city
models, Trans. GIS 16 (5) (2012) 681–690, http://dx.doi.org/10.1111/
j. 1467-9671.2012.01337.x.
[30] J.A. Jakubiec, C.F. Reinhart, A method for predicting city-wide electricity
gains from photovoltaic panels based on LiDAR and GIS data combined
with ho simulations, Solar Energy 93 (2013) 127–143, http://dx.doi.org/
10.1016/j.solener.2013.03.022.
[31] D.A. Jacques, J. Gooding, J.J. Giesekam, A.S. Tomlin, R. Crook, Methodology for
the assessment of PV capacity over a city region using low-resolution LiDAR
data and application to the City of Leeds (UK), Appl. Energy 124 (2014) 28–34,
http://dx.doi.org/10.1016/j.apenergy.2014.02.076.
[32] J. Liang, J. Gong, W. Li, A visualization-oriented 3D method for efficient
computation of urban solar radiation based on 3D-2D surface mapping,
Int. J. Geogr. Inf. Sci. 28 (4) (2014) 780–798, http://dx.doi.org/10.1080/
13658816.2014.880168.
 M. Bizjak et al. / Energy and Buildings 92 (2015) 195–203
[33] R. Storn, K. Price, Differential evolution – a simple and efficient heuristic for
global optimization over continuous spaces, J. Global Optimiz. 11 (4) (1997)
341–359, http://dx.doi.org/10.1023/A:1008202821328.
[34] J. Brest, S. Greiner, B. Boskovič, M. Mernik, V. Žumer, Self-adapting con-
trol parameters in differential evolution: a comparative study on numerical
benchmark problems, IEEE Trans. Evolut. Comput. 10 (6) (2006) 646–657,
http://dx.doi.org/10.1109/TEVC.2006.872133.
[35] D. Mongus, N. Lukač, B. Žalik, Ground and building extraction from LiDAR data
based on differential morphological profiles and locally fitted surfaces, ISPRS
J. Photogramm. Remote Sens. 93 (2014) 145–156, http://dx.doi.org/10.1016/
j.isprsjprs.2013.12.002.
[36] J. Heinzel, B. Koch, Exploring full-waveform LiDAR parameters for tree species
classification, Int. J. Appl. Earth Observ. Geoinf. 13 (1) (2011) 152–160,
http://dx.doi.org/10.1016/j.jag.2010.09.010.
[37] W.Y. Yan, A. Shaker, N. El-Ashmawy, Urban land cover classification using air-
borne LiDAR data: a review, Remote Sens. Environ. 158 (0) (2015) 295–310,
http://dx.doi.org/10.1016/j.rse.2014.11.001.
203
[38] D. Shepard, A two-dimensional interpolation function for irregularly-spaced
data, in: Proceedings of the 1968 23rd ACM National Conference, ACM
‘68, New York, USA, 1968, pp. 517–524, http://dx.doi.org/10.1145/800186.
810616.
[39] E. Mezura-Montes, J. Velázquez-Reyes, C.A.C. Coello, A comparative study of
differential evolution variants for global optimization, in: Proceedings of the
8th Annual Conference on Genetic and Evolutionary Computation, GECCO
‘06, New York, USA, 2006, pp. 485–492, http://dx.doi.org/10.1145/1143997.
1144086.
[40] N. Lukač, B. Žalik, GPU-based roofs’ solar potential estimation using
LiDAR data, Comput. Geosci. 52 (0) (2013) 34–41, http://dx.doi.org/10.1016/
j.cageo.2012.10.010.
[41] I. Reda, A. Andreas, Solar position algorithm for solar radiation appli-
cations, Solar Energy 76 (5) (2004) 577–589, http://dx.doi.org/10.1016/
j.solener.2003.12.003.