You are on page 1of 13

Journal of Building Engineering 43 (2021) 102525

Contents lists available at ScienceDirect

Journal of Building Engineering


journal homepage: http://www.elsevier.com/locate/jobe

A new metric for optimal visual comfort and energy efficiency of building
lighting system considering daylight using multi-objective particle
swarm optimization
Khairul Rijal Wagiman a, b, Mohd Noor Abdullah b, *, Mohammad Yusri Hassan c,
Nur Hanis Mohammad Radzi b
a
Industrial Training Institute Selandar, Jalan Batang Melaka, 77500 Selandar, Melaka, Malaysia
b
Green and Sustainable Energy (GSEnergy) Focus Group, Faculty of Electrical and Electronic Engineering, Universiti Tun Hussein Onn Malaysia (UTHM), 86400 Batu
Pahat, Johor, Malaysia
c
Centre of Electrical Energy Systems (CEES), School of Electrical Engineering, Faculty of Engineering, Universiti Teknologi Malaysia (UTM), 81310 Johor Bahru, Johor,
Malaysia

A R T I C L E I N F O A B S T R A C T

Keywords: In the past, an optimization-based control strategy of lighting in a building only focused on maximizing energy
Daylight-linked control efficiency in which the results of the visual comfort of occupants is decreased at the minimum level. Therefore, a
Energy savings visual comfort metric model that integrates the artificial lighting and daylighting is crucial to be developed, so
Multi-objective particle swarm optimization
that the model can be simultaneously treated with energy performance model in the control strategy for equal
Optimization-based control
Visual comfort metrics
performance of energy and visual comfort. This paper proposes a new visual comfort metric model called illu­
minance uniformity deviation index (IUDI) that reflects illuminance uniformity and the index is considered as an
objective function in the optimization-based control, based on a simulation-based study. The proposed IUDI is
optimized simultaneously with power demand using multi-objective particle swarm optimization (MOPSO) to
find the best compromise solution. The IUDI with MOPSO is implemented in an actual office room by considering
mixed artificial light and daylight. Based on the comparative results with the coefficient of variation of root mean
square error (CVRMSE), the proposed IUDI showed superior performance in two aspects: (1) technical ‒
improved 6% and 27% of visual comfort and energy savings, respectively and (2) economic ‒ reduced a simple
payback period of 28% and a net present value of 33%. The proposed IUDI is a great tool for a new visual comfort
metric that satisfies the EN 12464-1 (illuminance level is 500 lux and uniformity is 0.6). Moreover, it also has a
practical value to be implemented in green buildings.

implementing a daylight-linked control for the benefits of building


owners, occupants, and the environment. Apart from these strategies,
1. Introduction based on a recent study, modification of the building envelope (e.g.,
windows) can improve energy performance of lighting and visual
Buildings consume a significant amount of energy; for instance, comfort of occupants by considering daylighting [4].
buildings consumed around 40% of the total energy in the United States So far, lighting control in buildings has significantly attracted the
[1]. From that amount, 17% of the total electrical energy in commercial attention of researchers as this strategy can contribute to higher energy
buildings belongs to lighting, which is the second largest after others (e. savings [5], particularly when dealing with daylight-linked control
g., miscellaneous electric loads, process equipment, motors, and air [6–9]. The main goal of lighting control is to maximize energy savings
compressors) [2]. In the global context, fossil fuel is a prominent source and occupants’ visual comfort simultaneously. Visual comfort can be
of electricity, which contributed up to 80% in 2015 [3], and it is asso­ determined based on the light quantity and distribution [10]. Light
ciated with higher greenhouse gas (GHG) emissions. Therefore, based on quantity represents the amount of light on the horizontal plane, spe­
this fact, energy-efficient strategies in lighting systems should be cifically the working plane of the room, and this metric is commonly
considered, such as utilizing high energy-efficient lamps and

* Corresponding author.
E-mail address: mnoor@uthm.edu.my (M.N. Abdullah).

https://doi.org/10.1016/j.jobe.2021.102525
Received 1 September 2020; Received in revised form 3 April 2021; Accepted 11 April 2021
Available online 22 April 2021
2352-7102/© 2021 Elsevier Ltd. All rights reserved.
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

Nomenclature μo membership function for oth objective


v(t) current velocity of the particles
Symbols v(t+1) new velocity of the particles
c1 and c2 acceleration constants W total number of measurement points
di dimming level of luminaires at ith zone ωmax maximum and values of inertia weight
Ea artificial lighting illuminance matrix ωmin minimum values of inertia weight
Eav calculated average illuminance level x(t) current positions of the particles
Ed (t) daylight illuminance with respect to time matrix
x(t+1) new positions of the particles
Ei illuminance level of ith zone matrix
Ej,k illuminance value on the measurement point Acronyms
Ei,fd illuminance level when all luminaires in ith zone are fully 3D three-dimensional
dimmed matrix ANNs artificial neural networks
Em maintained average illuminance level BESs building energy systems
Emax maximum illuminance value of matrix ET (t) CC capital investment cost
ET (t) total distributed illuminance with respect to time matrix CD crowding distance
fomin lowest value of oth objective function CO convex optimization
fomax highest value of oth objective function COR competition over resources
Dmin lower bound of dimming level CVRMSE coefficient of the variation of root mean square error
Dmax upper bound of dimming level EC energy consumption
gbest (t) global best positions ECVC energy consumption and visual comfort
I the total number of control zones ER external repository
iterj current iteration number ES extremum seeking
itermax maximum number of iterations GHG greenhouse gas
k number of non-dominated solutions HVAC heating, ventilation, and air conditioning
Nav normalized illuminance values of average point ID illuminance deviation
Nj,k normalized illuminance values (j,k) point IUDI illuminance uniformity deviation index
LED light-emitting diode
pbest(t) local best positions
LP linear programming
PD power demand
MOPSO multi-objective particle swarm optimization
r1 and r2 random numbers between [0,1]
NPV net present value
ti 1-h time interval of the working hours
NSGA II non dominated sorting genetic algorithm II
Uh higher value set point of uniformity
PSO particle swarm optimization
Uo calculated illuminance uniformity
Uo,max maximum value of Uo SPP simple payback period
UD uniformity deviation
Uo,min minimum illuminance uniformity
μk linear membership function

measured in lux, which refers to illuminance. Similarly, light distribu­ utilized, including particle swarm optimization (PSO) [27,28] and
tion, known as illuminance uniformity, also measures the light on the modified competition over resources (COR) algorithm [29]. In a recent
working plane and focuses on its illuminance level distribution. The study [8], a combination of LP and artificial neural networks (ANNs)
prominent quantity and distribution of the light metrics are represented was proposed to handle a single-objective optimization-based control
by the average illuminance level (Em) and minimum illuminance uni­ for minimizing the dimming levels of luminaires while satisfying Em and
formity (Uo,min), respectively, where both illuminance-based metrics are Uo,min. Meanwhile, in a multi-objective approach, Madias et al. [30]
specified in the European Standard EN 12464-1 [11]. According to the applied the coefficient of the variation of root mean square error
EN 12464-1, the minimum values of Em and Uo,min for offices are 500 lux (CVRMSE) as an indicator to measure the Uo in a room. A lower value of
and 0.6, respectively. Higher values of the calculated average illumi­ CVRMSE produced a higher value of Uo. CVRMSE was considered as the
nance level (Eav) and illuminance uniformity (Uo) in a room environ­ objective function and treated simultaneously with the dimming levels
ment offer better visual comfort, and this environment is also good for of luminaires. The bi-objective function was optimized using
health, well-being, and sleep quality [12]. non-dominated sorting genetic algorithm-II (NSGA-II). However, the
In lighting control, the dimming level of light-emitting diode (LED) study only considered illuminance distribution from artificial lighting;
luminaire is proportional to its output, meaning that by minimizing the thus, daylighting was not considered and the CVRMSE model required
dimming levels of LED luminaires, energy consumption (EC) is also further improvement to obtain better results for Uo.
minimized [13–16]. Lighting control strategies based on the dimming Based on the current literature survey for optimization-based control
levels of luminaires can be grouped based on their objective functions: strategies, most of the works considered a single-objective function that
single- and multi-objective approaches. In the single-objective minimizes the dimming levels of artificial lighting with illuminance-
approach, only one objective is considered, which is the summation of based metrics (e.g., illuminance level and uniformity) as the con­
the dimming levels of luminaires to be minimized that satisfies the straints. In a single optimization-based strategy, the illuminance-based
constraints of Em and Uo,min. Majority of the researchers have used metrics are achieved at the minimum levels, reflecting the minimum
conventional optimization methods to find the optimum solution of the level of visual comfort of occupants as only EC is minimized. Therefore,
dimming levels of luminaires (i.e., single objective) [17], including a multi-objective optimization-based strategy is required to minimize EC
convex optimization (CO) [18,19], linear programming (LP) [13, and maximize visual comfort simultaneously. Based on the literature
20–22], iterative method [23,24], and extremum seeking (ES) [25,26]. survey, only a single work utilized a multi-objective optimization
Apart from that, meta-heuristic optimization methods have also been technique in the optimization-based control strategy by considering EC

2
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

and visual comfort of occupants (i.e., illuminance uniformity) as the axis (j × k), and this idea is presented in Fig. 1.
objective functions [30]. However, daylighting was not considered in Based on Fig. 1, the daylight illuminance levels on the working plane
the problem formulation of their work and the CVRMSE model needed with respect to time can be formulated in the matrix function denoted
further research to improve the illuminance uniformity in the room. byEd (t), as described in the following equation:
Therefore, a new mathematical model for representing illuminance ⎡ ⎤
E1,1 E1,2 ⋯ E1,k
uniformity by considering daylighting is needed in the ⎢ E2,1 E2,2 ⋯ E2,k ⎥
optimization-based control to improve the visual comfort of occupants Ed (t) = ⎢
⎣⋮
⎥ (1)
⋮ ⋱ ⋮ ⎦
and achieve the optimum level of EC simultaneously. Ej,1 Ej,2 ⋯ Ej,k
The inclusion of daylighting in optimization-based control can
reduce the EC of artificial lighting because the illuminance levels The matrix size of Ed (t) is equal to the number of measurement
contributed by daylight in a room reached thousands of lux (e.g., points with the arrangement of (j × k).
1000–3500 lux), particularly for the area near to a window than the
illuminance levels contributed by artificial lighting in the range of 2.2. Mathematical formulation of artificial lighting model
300–700 lux around the room. On the other hand, the illuminance levels
at the area far from the window are much lower than the illuminance Artificial lighting is used to provide illumination in a room so that
levels at the area close to the window. Hence, the illuminance unifor­ human activities and tasks can be performed effectively and comfort­
mity in the room is significantly lower, and it is challenging to sustain ably. Artificial lighting can be modeled mathematically based on its
the visual comfort of occupants when considering Uo as a design output (lumen), which is measured on the working plane with specified
parameter or metric. Moreover, the application of CVRMSE as an indi­ measurement points (j × k). This concept can be illustrated based on the
cator for Uo only considers artificial lighting [30]. Therefore, this paper three-dimensional (3D) model as outlined in Fig. 2.
proposed a new mathematical model considering daylighting called the Based on Fig. 2, each illuminance value (Ej,k) on the measurement
illuminance uniformity deviation index (IUDI) to simultaneously point from artificial lighting can be formed into a matrix function called
improve Uo and EC. The proposed IUDI is incorporated with artificial lighting illuminance matrix (Ea) and it can be expressed as:
multi-objective particle swarm optimization (MOPSO) to generate a set ⎡ ⎤
of optimal dimming level vectors with minimized EC and maximized E1,1 E1,2 ⋯ E1,k
⎢ E2,1 E2,2 ⋯ E2,k ⎥
visual comfort metric (Uo). Special attention is paid to consider both Ea = ⎢ ⎥ (2)
⎣⋮ ⋮ ⋱ ⋮ ⎦
illuminance-based metrics (i.e., Eav and Uo) as design parameters (fully
Ej,1 Ej,2 ⋯ Ej,k
comply with the EN 12464-1) and both illuminance sources (i.e., arti­
ficial lighting and daylighting). MOPSO is chosen because the algorithm In some cases, a lighting control scheme in a building adapted the
demonstrated superior performance for solving multi-objective building zoning control scheme, in which the luminaires are grouped into ith
energy systems (BESs) problems, for example, in Refs. [31,32]. zone. Therefore, Ea is a summation of the illuminance matrices of ith
The main purpose of this paper is to develop a new visual comfort zone (Ei), which can be represented as:
metric model called the IUDI by considering daylight-responsive control

I
and incorporating the model with MOPSO. The scopes of this study Ea = Ei (3)
considered two illuminance-based visual comfort metrics ofEav and Uo i=1
(comply with the EN 12464-1) and a simulation-based study. Other vi­
sual comfort metrics (e.g., glare) and a small-scale application of the Ei = di Ei,fd (4)
proposed IUDI with a MOPSO-based control strategy will be considered
in future research. Three contributions of this research are: where I is the total number of control zones, di is the dimming level of
luminaires at ith zone and computed by the optimizer, and Ei,fd is the
• Formulation of a new visual comfort metric model (IUDI) by illuminance level matrix when all luminaires in ith zone are fully dim­
considering daylighting to improve the performance of visual com­ med and the rest of the luminaires at the zones are set at zero dimming
fort and energy. (i.e., switched off), where it can be rearranged as a matrix function as
• Development of a multi-objective optimization-based control strat­ follows:
egy of lighting systems by treating energy and visual comfort ⎡
E1,1 E1,2 ⋯ E1,k

simultaneously as the main objective functions. ⎢ E2,1 E2,2 ⋯ E2,k ⎥
• The proposed model (IUDI) provides superior performance for en­ Ei,fd = ⎢
⎣⋮
⎥ (5)
⋮ ⋱ ⋮ ⎦
ergy and visual comfort performances than the existing model Ej,1 Ej,2 ⋯ Ej,k
(CVRMSE).
The total distributed illuminance matrix (ET (t)) in the room is the
The rest of this paper is organized as follows. Section 2 covers the
formulation of mathematical models. In Section 3, the formulation of
energy consumption and visual comfort (ECVC) problems is described.
Section 4 presents the development of the proposed MOPSO algorithm
for solving ECVC problems. Section 5 consists of the simulation results
and discussion. Finally, the conclusion of the study and future work
recommendations are presented in Section 6.

2. Mathematical models

2.1. Mathematical formulation of daylighting model

Daylight penetration through a window of a building can be modeled


in a mathematical function by measuring the illuminance levels on the
horizontal plane, specifically the working plane with a specific number Fig. 1. Three-dimensional representation of distributed illuminance on the
of measurement points with j points in the x-axis and k points in the y- working plane from daylighting.

3
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

formulation, it is clearly demonstrated that the proposed IUDI model is a


complete and useful model for representing Uo and Eav on the working
plane in the room. The mathematical expression of the proposed IUDI
model is given as:
( ⃒ ⃒/ ) ( )
K ⃒
∑ J ∑
Nav − Nj,k ⃒ Uh − Uo
IUDI = W × (9)
j=1 k=1
Nav Uh
⏟̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅⏟ ⏟̅̅̅̅̅̅̅̅⏞⏞̅̅̅̅̅̅̅̅⏟
UD
ID

where J and K are the total number of measurement points in row and
column, respectively, of matrix ET (t), W = J × K; Nav and Nj,k are
normalized illuminance values of average and (j,k) point, respectively;
and Uh is the higher value set point of uniformity and the value is
rounded up to 1 decimal value (i.e., either 0.7, 0.8, 0.9, or 1), where Uh
Fig. 2. Three-dimensional representation of distributed illuminance on the
working plane from artificial lighting.
should be higher than the maximum value of Uo (Uo,max ). For example, if
Uo,max is 0.72, then Uh is set at 0.8. The Nav can be calculated by dividing
the average illuminance value (Eav) with the maximum illuminance
combination of Ea and Ed (t), and it can be represented as a mathematical
value (Emax) of matrix ET (t), as given in Eq. (10), and Nj,k can be
function:
determined by dividing (j,k) point illuminance value (Ej,k) with Emax of
ET (t) = Ea + Ed (t) (6) matrix ET (t), as presented in Eq. (11):
It can be deduced that the values in matrix ET (t) fully depend on two Eav
Nav = (10)
components, namely di (i.e., in matrixEa ) andEd (t). Only di can be Emax
controlled to satisfy occupants’ comfort in the room. Thus, the value of
di is considered as a set of decision variables in an optimization-based Nj,k =
Ej,k
(11)
control strategy. Emax
Based on Eq. (6), visual comfort metrics, namely average illuminance
level (Eav ) and illuminance uniformity (Uo) can be calculated. Both 3. Problem formulation
metrics are specified in the EN 12464-1. Eav and Uo can be represented as
mathematical functions: 3.1. Objective functions

Eav = avg[ET (t)] (7)


3.1.1. Energy consumption of artificial lighting
Emin As highlighted in Section 1, reducing the dimming levels of LED lu­
Uo = (8) minaires results in reduced energy consumption of lighting. Thus, the
Eav
objective function is formulated as:
where Emin = min[ET (t)]. ∑
K
Min FD = di (12)
2.3. Proposed illuminance uniformity deviation index
i=1

where FD is the objective function, with the summation of di and di is the


Uo is one of the visual comfort metrics that corresponds to the quality dimming levels of the luminaires for ith zone.
of distributed illuminance on the working plane [10], which is
mentioned in the EN 12464-1. Higher value of Uo indicates better visual 3.1.2. Proposed illuminance uniformity deviation index
environment. However, the EN 12464-1 does not provide any specific The proposed IUDI (Eq. (9)) is considered as an objective function
mathematical model to determine Uo when the illuminance distribution (FU) to be minimized, in which a small value means higher value of Uo.
on the working plane is influenced by di (i.e., for artificial lighting) (Eq. The FU is to represent visual comfort of occupant in a room/space. The
(2)) and daylighting (Eq. (1)), where the value changes with respect to mathematical expression of the proposed IUDI is given as:
time. Meanwhile, in Ref. [30], the CVRMSE only considered the
dimming levels of the artificial lighting model and did not include the Min FU = Min IUDI (13)
daylighting model. The visual comfort metric model is crucial when
considering a multi-objective optimization-based control in the lighting
3.2. Constraints
to improve energy performance and visual comfort, as well as to fully
satisfy the EN 12464-1. Thus, in this paper, a new illuminance-based
3.2.1. Luminaire’s dimming limits
index called illuminance uniformity deviation index (IUDI) model is
The allowable range of the dimming levels of luminaires of ith zone
proposed. The proposed model consists of two terms. The first term of
(di ) is:
the proposed IUDI model is called illuminance deviation (ID). The
concept of ID is the summation of different calculated values of illumi­ Dmin ≤ di ≤ Dmax (14)
nance between an average and each measurement point, and then
divided by the average illuminance value, where all the values are where Dmin and Dmax are the lower and upper bounds, respectively.
normalized. The summation value is divided by the total number of
measurement points (W). By considering all measurement points on the 3.2.2. Average illuminance level
measured plane into the calculation, the effectiveness of ID for repre­ The Eav can be calculated using Eq. (7). Meanwhile, the maintained
senting the illuminance levels across the room is increased. Meanwhile, average illuminance level (Em ) is commonly based on standards (e.g., EN
the second term of the proposed IUDI model is known as uniformity 12464-1). The constraint is:
deviation (UD). UD has different values of higher uniformity set-point
Eav ≥ Em (15)
with calculated uniformity, and then is divided by the values of higher
uniformity set-point. Finally, ID is multiplied with UD. Based on this

4
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

3.2.3. Illuminance uniformity


Uo can be calculated using Eq. (8). On the other hand, the minimum
value of illuminance uniformity (Uo,min) is according to standards. The
constraint is:
Uo ≥ Uo,min (16)

4. Proposed multi-objective particle swarm optimization


algorithm for optimal energy consumption and visual comfort
problems

4.1. Particle swarm optimization

PSO is a population-based optimization technique inspired by animal


social behavior (e.g., fish schooling and bird flocking), and the approach
was introduced by Kennedy and Eberhart [33]. Individuals of a group of
animals are called particles and the group is called a swarm. Initially, the
positions of particles are generated randomly within a search space. The
particles fly from one position to other positions based on the specified
velocity and the velocity is updated continuously until the global best
position (gbest) is reached, which is the optimum solution. The new
velocity (v(t+1) ) and position (x(t+1) ) update processes are [34]:
( ) ( )
v(t+1) = ωv(t) + c1 r1 pbest(t) − x(t) + c2 r2 gbest(t) − x(t) (17) Fig. 3. Pareto-optimal front and dominated solutions with two objectives.

x(t+1) = x(t) + v(t+1) (18) 4.4. Crowding distance

where v(t) and v(t+1) are the current and new velocity of the particles, Crowding distance (CD) approach is used to maintain the number of
respectively; c1 and c2 are acceleration constants; r1 and r2 are random non-dominated solutions within a specified size and to produce a well-
numbers between [0,1]; pbest(t) and gbest(t) are the local and global best distributed Pareto-optimal front. The value of CD for every single non-
positions, respectively; and x(t) and x(t+1) are the current and new po­ dominated solution in the ER needs to be calculated and the lowest
sitions of the particles, respectively. For increasing the performance of value is deleted. The value of CD for rth non-dominated solution can be
the algorithm in terms of fast convergence, the linear decreasing inertia computed using the following equation [35]:
weight method can be used [32]: p
∑ fo (j + 1) − fo (j − 1)
(
ωmax − ωmin
) CDr = (22)
ω = ωmax − × iterj (19) o=1
fomax − fomin
itermax
where j is jth iteration and o is the number of objective functions, and
where iterj is the current iteration number, itermax is the maximum
fomin and fomax are the lowest and highest values of oth objective function,
number of iterations, andωmax and ωmin are the maximum and minimum
respectively.
values of inertia weight, respectively.
4.5. Proposed multi-objective particle swarm optimization (MOPSO)
4.2. Pareto-optimal front algorithm for energy consumption and visual comfort problems

Pareto-optimal front is a set of optimal solutions of a feasible solution The simulation-based lighting system optimization framework is
region, which is also called non-dominated solutions. Fig. 3 presents the illustrated in Fig. 4. Initially, the daylighting (Ed (t)) for each hour of the
relation of non-dominated solutions and dominated solutions in two
objective functions. By referring to the figure, there are two objectives (i.
e., f1(x) and f2(x)) and both objectives are considered to be minimized,
where non-dominated solutions can be determined based on several
rules. A feasible solution (xs ) dominates another feasible solution (xs+1 )
if both conditions are satisfied:
fo (xs ) ≤ fo (xs+1 ) for all objectives (20)

fo (xs ) < fo (xs+1 ) for at least one objective (21)

where o = 1, 2, …, k. Then, xs is a non-dominated solution if another


solution does not dominatexs .

4.3. External repository

Non-dominated solutions are stored in the external repository (ER)


with a specified size. These solutions are updated continuously during
optimization process by inserting and deleting them. The deleting pro­
cess occurs when the non-dominated solutions exceed the specified size
based on crowding distance approach. Fig. 4. Simulation-based lighting system optimization framework.

5
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

working hours (8:00–17:00 h) and artificial lighting (Ea) illuminance


matrices are developed, as discussed extensively in Sections 2.1 and 2.2,
respectively. Next, the MOPSO algorithm is applied in MATLAB plat­
form throughout the working hours. Based on Fig. 4, ti represents 1-h
time interval of the working hours, which is initially set at 8 (i.e.,
8:00 h). Detailed explanation of the optimization-based control using
MOPSO is presented in the following paragraph. After the results of the
Pareto-optimal front (MOPSO results) are obtained, the best compro­
mise solution of the dimming levels of luminaires of the ith zone (di) is
determined. The discussion of the best compromise solution is presented
in the following section. Finally, from the results of di, the value of Uo is
calculated and an isolux diagram is plotted using DIALux.
The parameter settings and their constraints to perform the
optimization-based control using MOPSO are shown in Table 1.
The procedures of the proposed MOPSO to generate Pareto-optimal
front for solving ECVC problems are as follows:

(1) Load illuminance data (i.e., artificial lighting and daylighting).


(2) Set the parameters and constraints (Table 1).
(3) Randomly generate the initial position of a group of n particles
(x(t) ) within the bounds in Dmin and Dmax for all dimensions. The
position of particle (x) represents the dimming level of luminaires
of ith zone (di ).
(4) Calculate Eav and Uo using Eqs. (7) and (8), respectively, for each
particle.
(5) Check whether Eav and Uo of each particle satisfy the constraints
(i.e., Eqs. (15) and (16), respectively). If both Eav and Uo do not
satisfy the constraints, randomly generate the position of the
particles for those particles that do not satisfy the constraints and
repeat step 3 to this step until all particles fully satisfy the
constraints.
(6) Evaluate the objective function values of all particles and update
pbest and gbest.
(7) Update the ER by inserting the non-dominated solutions into the
ER and remove the dominated solutions from the ER.
(8) Calculate the new velocity (v(t+1) ) and position for each particle
Fig. 5. Flowchart of the proposed MOPSO algorithm for solving
(x(t+1) ) using Eqs. (17) and (18), respectively. ECVC problems.
(9) If the stopping criterion (i.e., maximum iteration) is satisfied,
terminate the process. Otherwise, repeat steps 4 to 8.
( )
Max μk (23)
The flowchart of the proposed MOPSO algorithm for solving ECVC
problems of an indoor lighting system is outlined in Fig. 5. ∑
N
μko
μk = o=1
(24)

M ∑ N
4.6. Best compromise solution μko
k=1 o=1

In this paper, a well-known approach for determining the best ⎧


1, fo ≤ fomin
compromise solution is used, which is fuzzy set, and this approach is ⎪



based on the maximum value of a linear membership function (μk ), and ⎨ f max − f
(25)
o
μo = o
, fomax < fo < fomin
the mathematical expressions are [35]: ⎪
⎪ fomax − fomin



0, fo ≥ fomax
Table 1
Parameter settings for MOPSO and their constraints.
where μo is the membership function for oth objective and k is the
Parameters number of non-dominated solutions.
itermax 300
ωmin 0.4 5. Simulation results and discussion
ωmax 0.9
c1 and c2 2 5.1. Case study
Number of particles, n 30
Dimension (i.e., number of zones) 12 To evaluate the effectiveness of the proposed IUDI incorporated with
v(t) 0 MOPSO, an office room of the Industrial Training Institute Selandar,
Constraints Melaka, Malaysia is considered. This building is stand-alone; hence,
Dmin 0
there are no urban obstacles/neighboring buildings. The room is
Dmax 1
Em 500 lux
attached with three windows at one side, and a single window and two
Uo,min 0.6
doors on the other side. In this case, the standard glass for windows (in

6
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

DIALux) is considered and no shading devices are used. The floor plan Table 2
and luminaire grid array are shown in Fig. 6. The design and simulation Parameters for simulating in DIALux.
parameters in DIALux are referred to the EN 12464-1 and listed in Parameters
Table 2.
Location Longitude: 102.4◦ , Latitude: 2.4◦
North alignment 189◦
5.2. Baseline Date October 5, 2018
Period 8:00 to 17:00 h (1-h interval) - office hours
Sky condition Clear
The simulation results of the case study that considers daylighting Luminaire
only (condition 1) and a combination of daylighting and artificial Type Philips BBS562 1xLED35S/840 AC-MLO (LED)
lighting with all luminaires are fully dimmed, which is 100% dimming Number 35
(condition 2), are tabulated in Table 3. Based on the table, for condition Lumen 3500 lm
Power 34 W
1, at 8:00 h, the lowest Eav of 209 was recorded. Meanwhile, the lowest
Room dimensions 8 m × 20 m × 2.7 m (W × L × H)
Uo of 0.190 was recorded at 16:00 h. The values of visual comfort Window dimensions 2.43 m × 1.65 m (W × H)
metrics (i.e., Eav and Uo) varied depending on time. Meanwhile, for Number of measurement points 108 (9 × 12)
condition 2, the highest value of Eav of 1017 lux was recorded at 16:00 h Working plane height 0.75 m
and the lowest value of Uo of 0.52 was recorded at both 15:00 and 16:00 Surface reflection factors
Ceiling, ρc 0.7
h. Only both Eav and Uo values at 8:00 h fully satisfied the recommended Wall, ρw 0.5
values and the rest of the times only satisfied the recommended Eav of Floor, ρf 0.2
the EN 12464-1. Thus, a multi-objective optimization-based control Em 500 lux
strategy is required to optimize the dimming levels of luminaires by Uo,min 0.6
trading-off two competing objective functions (i.e., power demand (PD)
and visual comfort metrics) simultaneously. In this paper, condition 2 is
The isolux diagram at the room at 16:00 h obtained by minimizing
considered as the baseline with PD of 1.19 kW for each hour and the total
the proposed IUDI (i.e., maximum Uo) is illustrated in Fig. 8. It can be
PD of 11.9 kW throughout working hours. Fig. 7 depicts the isolux dia­
seen that the illuminance distribution across the room improved
gram at 16:00 h by considering condition 2.
significantly than the results in Fig. 7, in which most of the area of the
room has the value of illuminance around 500 lux. It means that the
5.3. Optimization-based control illuminance distribution is more uniform (i.e., higher uniformity).
The comparison of the baseline and the proposed IUDI in terms of PD

To evaluate the performance and effectiveness of the proposed IUDI and Uo by minimizing d individually is depicted in Fig. 9. It is clearly

as an objective function, the proposed FU = IUDI (Eq. (13)) and FD = d shown that the PD contributed by the proposed IUDI (green) is signifi­
(Eq. (12)) are minimized individually and simultaneously, and both cantly lower than the baseline (red) throughout working hours. The total
considerations are subject to Eqs. (14)–(16). Moreover, the proposed FU PD of 3.983 kW was recorded throughout working hours and the total
is also compared with CVRMSE [30] in terms of PD and Uo. energy savings of 67% was achieved compared to the baseline (11.9
kW). Meanwhile, for Uo, most of the times, the Uo contributed by the
5.3.1. Best solutions proposed IUDI (blue) is higher than the baseline (yellow), except at 8:00
Firstly, the proposed FU and FD are minimized individually. The h. The average of Uo contributed by the proposed IUDI throughout
average computational time for ten runs was 3.54 s. Table 4 shows the working hours was 0.6, and 9% improvement was achieved compared to
best solutions of the objective functions for selected times. It can be seen the average of Uo of the baseline (0.55).

that the lowest PD was 0.335 kW at 17:00 h with the minimum of d = The comparison of the baseline and the proposed IUDI in terms of PD
3.188, and the highest value of Uo was 0.75 at 8:00 h with the minimum and Uo by minimizing the IUDI individually is illustrated in Fig. 10. It
of IUDI = 0.009. Moreover, the proposed IUDI is also capable to produce can be seen that the PD contributed by the proposed IUDI (green) is lower
higherEav . By minimizing FD , the total PD of 3.983 kW and the average Uo than the baseline (red) throughout working hours. The total PD of 9.366
of 0.602 were produced throughout working hours. Meanwhile, by kW was recorded and the total energy savings of 21% was attained
minimizing FU , the total PD of 7.259 kW and the mean Uo of 0.69 were compared to the baseline. Meanwhile, for Uo, it is clearly shown that at
recorded throughout working hours.

Fig. 6. Floor plan, light sensor positions, and their control zones of the room.

7
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

Table 3
Simulation results for conditions 1 and 2 without optimization-based control.
Condition Parameter 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00

1 Eav 209 232 360 357 348 344 355 372 382 344
Uo 0.239 0.229 0.223 0.223 0.220 0.201 0.194 0.191 0.190 0.193
2 Eav 844 958 995 992 983 980 990 1008 1017 979
Uo 0.63 0.58 0.56 0.55 0.54 0.54 0.53 0.52 0.52 0.53

Note:
Condition 1 considered daylighting only.
Condition 2 combined daylighting and artificial lighting with fully dimming (100%) for all luminaires.

Fig. 7. Isolux diagram at the room at 16:00 h (the lowest Uo) by considering condition 2.

Table 4

Best solutions for dand IUDI for selected times.
Result 8:00 12:00 16:00 17:00
∑ ∑ ∑ ∑
Best d Best IUDI Best d Best IUDI Best d Best IUDI Best d Best IUDI

d1 0.585 0.887 0.446 0.776 0.270 0.499 0.270 0.406


d2 0.347 0.405 0.052 0.268 0.000 0.195 0.000 0.334
d3 0.524 0.675 0.260 0.501 0.260 0.444 0.260 0.393
d4 0.223 0.517 0.000 0.301 0.000 0.288 0.000 0.269
d5 0.291 0.684 0.270 0.484 0.270 0.410 0.270 0.403
d6 0.328 0.558 0.000 0.275 0.000 0.188 0.061 0.221
d7 0.416 0.931 0.655 1.000 0.686 1.000 0.600 1.000
d8 0.836 1.000 0.619 0.921 0.662 0.905 0.492 0.863
d9 0.454 0.888 0.407 0.775 0.259 0.612 0.230 0.567
d10 0.000 0.302 0.000 0.000 0.000 0.000 0.097 0.160
d11 0.250 0.651 0.351 0.781 0.329 0.787 0.389 0.749
d12 0.807 0.974 0.710 0.919 0.734 0.919 0.518 0.890

d 5.060 8.472 3.772 7.001 3.470 6.245 3.188 6.256
IUDI 0.046 0.009 0.087 0.038 0.084 0.040 0.079 0.029
PD (kW) 0.548 0.869 0.404 0.724 0.381 0.651 0.335 0.648
Uo 0.604 0.750 0.601 0.685 0.601 0.680 0.606 0.702
Eav (lux) 501 673 560 732 582 727 520 688

Fig. 8. Isolux diagram at the room at 16:00 h by minimizing the proposed IUDI with MOPSO (maximum Uo).

8
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

Fig. 9. Comparison of the baseline and the proposed IUDI in terms of PD and Uo

by minimizing d.

Fig. 12. Pareto-optimal front at 16:00 h.

thus, it can be said that most of the solutions produced lower Uo than at
8:00 h. Generally, the distribution of illuminance across a room depends
on daylighting and time, which are influenced by the position of the sun
and light intensity. Early in the morning, the illuminance levels are
lower and the illuminance distribution is more uniform (i.e., referring to
illuminance uniformity) compared to at noon and in the evening (refer
to Table 3). However, MOPSO generates the optimum dimming levels to
achieve better results for two competing objective functions.
The best compromise solution is extracted from the generated
Pareto-optimal front using fuzzy-based approach and the results are
shown in Table 5. It can be seen that at 17:00 h, the lowest PD of 0.435

kW was recorded with the minimum value of d = 4.11 and at 8:00 h,
Fig. 10. Comparison of the baseline and the proposed IUDI in terms of PD and the highest Uo of 0.725 was achieved with the minimum value of IUDI =
Uo by minimizing the IUDI. 0.017.
Fig. 13 depicts the isolux diagram across the room at 16:00 h for the
all times, the Uo contributed by the proposed IUDI (blue) is significantly best compromise solution. It is clearly shown that the illuminance dis­
higher than the baseline (yellow). The mean of Uo contributed by the tribution across the room improved than the results in Fig. 7, in which
proposed IUDI throughout working hours was 0.69 and it showed 20% most of the area of the room has the value of illuminance around 500
improvement than the mean of Uo of the baseline. lux, and it also means that the illuminance distribution is more uniform.
Fig. 14 illustrates the comparison of the baseline and the proposed
5.3.2. Best compromise solution IUDI for the best compromise solution in terms of PD and Uo. It is clearly
In this section, the proposed objective function FU and FD are mini­ shown that the proposed IUDI (green) always produced lower PD than
mized simultaneously using MOPSO. The average computational time the baseline (red) throughout working hours. The proposed IUDI pro­
for ten runs was 21.25 s. Figs. 11 and 12 show the Pareto-optimal front duced the total PD of 5.047 kW and obtained 58% energy savings
of MOPSO at 8:00 and 16:00 h, respectively. According to Fig. 10 that compared to the baseline. Meanwhile, the proposed IUDI (blue) ach­
shows the Pareto curve generated by MOPSO, it can be interpreted that ieved significant Uo throughout working hours than the baseline (yel­
most of the solutions provided higher Uo. Differently, in Fig. 12, the span low). The mean of Uo of 0.661 was contributed by the proposed IUDI
of the Pareto curve is much narrower than the Pareto curve in Fig. 11; throughout working hours and attained 17% improvement compared to
the mean of Uo of the baseline.

5.3.3. Comparison with other model


In this section, the CVRMSE is adapted as an objective function [30]
and denoted as FCV to represent Uo. Firstly, FCV is minimized individually
to obtain the best value of CVRMSE, which corresponds to maximizing
Uo. Secondly, FCV and FD are minimized simultaneously and can be
presented mathematically as minimize [FD , FCV ]. It is then optimized
using the same MOPSO, which is used in the previous section. As
highlighted in Section 1, Ref. [30] applied the CVRMSE as an objective
function and the study did not consider daylight. Thus, in this research,
daylighting is taken into account in the CVRMSE. Finally, the results of
the CVRMSE are compared with the proposed IUDI in terms of PD and Uo.
The best solutions of CVRMSE (individually minimized FCV ) for
selected times are shown in Table 6. It can be seen that at 8:00 h, the
highest Uo of 0.7 was achieved with the highest PD of 0.999 kW was
recorded. The CVRMSE produced the average of Uo of 0.64 and the total
PD was 8.58 kW throughout working hours.
Fig. 11. Pareto-optimal front at 8:00 h. By referring to Table 4 (column of the best IUDI) and Table 6, the

9
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

Table 5

Best compromise solutions for dand IUDI.
Result 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00

d1 0.558 0.619 0.653 0.614 0.550 0.572 0.527 0.378 0.352 0.270
d2 0.280 0.192 0.192 0.141 0.104 0.130 0.127 0.058 0.101 0.096
d3 0.380 0.372 0.390 0.368 0.299 0.376 0.334 0.266 0.262 0.260
d4 0.363 0.167 0.164 0.090 0.051 0.095 0.124 0.054 0.118 0.128
d5 0.480 0.320 0.388 0.340 0.291 0.326 0.372 0.272 0.272 0.270
d6 0.183 0.108 0.138 0.095 0.073 0.128 0.129 0.096 0.082 0.123
d7 0.625 0.710 0.813 0.776 0.720 0.808 0.864 0.774 0.795 0.771
d8 0.682 0.736 0.783 0.731 0.685 0.787 0.745 0.746 0.703 0.675
d9 0.597 0.574 0.683 0.564 0.537 0.605 0.532 0.383 0.440 0.336
d10 0.043 0.026 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
d11 0.384 0.381 0.474 0.502 0.466 0.534 0.590 0.518 0.518 0.535
d12 0.682 0.679 0.819 0.705 0.752 0.791 0.714 0.627 0.826 0.647

d 5.258 4.883 5.497 4.927 4.529 5.152 5.058 4.170 4.470 4.110
IUDI 0.017 0.039 0.045 0.057 0.065 0.057 0.060 0.070 0.059 0.049
PD (kW) 0.548 0.513 0.576 0.515 0.477 0.540 0.527 0.441 0.475 0.435
Uo 0.725 0.678 0.668 0.652 0.640 0.653 0.650 0.635 0.648 0.663
Eav (lux) 501 594 665 629 599 630 633 605 632 573

Fig. 13. Isolux diagram at the room at 16:00 h for the best compromise solution.

Table 6
Best solutions of the CVRMSE for selected times.
Result 8:00 12:00 16:00

d1 1.000 1.000 0.656


d2 0.705 0.409 0.197
d3 0.910 0.898 0.741
d4 0.792 0.458 0.237
d5 0.860 0.782 0.784
d6 0.638 0.470 0.172
d7 0.989 1.000 1.000
d8 1.000 1.000 1.000
d9 0.990 0.871 0.721
d10 0.160 0.000 0.000
d11 0.700 0.837 0.840
d12 1.000 1.000 1.000
CVRMSE 0.183 0.383 0.390
Fig. 14. Comparison of the baseline and the proposed IUDI model for the best
Uo 0.700 0.621 0.636
compromise solution in terms of PD and Uo. PD (kW) 0.999 0.900 0.764

comparison of the proposed IUDI and CVRMSE plotted for 8:00, 12:00,
and 16:00 h is illustrated in Fig. 15. It can be seen that the proposed IUDI energy savings of 15%. Hence, it is demonstrated that the proposed IUDI
(red) produced higher Uo with lower PD than the CVRMSE (blue) for all has superior performance in individual minimization of objective
times displayed in the figure. At 8:00 h, the proposed IUDI contributed functions with higher visual comfort (Uo) and energy savings.
higher Uo of 0.75 (i.e., the highest value) than the CVRMSE with Uo of The results of the best compromise solution by minimizing [FD ,
0.7, with around 7% improvement for the proposed IUDI in comparison FCV ]simultaneously throughout working hours are presented in Table 7.
to the CVRMSE model. At the same time, the PD produced by the pro­ It can be observed that the highest Uo of 0.668 and PD of 0.723 kW were
posed IUDI of 0.869 kW was lower than the CVRMSE model (close to 1 recorded at 8:00 h. This model produced the mean of Uo of 0.62 and the
kW), with around 13% energy savings of the proposed IUDI in com­ total PD of 6.904 kW throughout working hours.
parison to the CVRMSE. Overall, throughout working hours, the per­ Based on Tables 5 and 7, the comparison of the proposed IUDI and
formance of the proposed IUDI achieved 7% improvement of Uo and CVRMSE for the best compromise solution plotted for 8:00, 12:00, and
16:00 h is presented in Fig. 16. It can be seen that for all times, the

10
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

considered as an input to the controller (Arduino). A dimmable LED


driver can control the dimming levels of LED luminaires, where the
signal comes from the controller.
In a large-scale application, a lighting system is one of the sub-
systems of building management system (BMS), where the controller
and dimmers are already installed in the building. The sub-lighting
system is integrated with BMS through the BACnet protocol [37], as
illustrated in Fig. 17. Thus, the proposed IUDI-MOPSO algorithm can be
programmed to the existing controller of the sub-lighting system (red
rectangle in Fig. 17).

5.5. Economic analysis

Two widely used economic performance metrics are the simple


payback period (SPP) and the net present value (NPV). SPP is the period
required to reclaim the capital investment cost (CC) with respect to cost
savings, and it is a simple economic analysis method. SPP can be
described in Eq. (26) [38]. NPV is the difference of summation of the
Fig. 15. Comparison of the proposed IUDI and CVRMSE by minimizing the present values (PVs) of k years with CC, and the formula is given in Eq.
proposed IUDI and CVRMSE individually for selected times.
(27) [39].

proposed IUDI achieved higher Uo and lower PD than the CVRMSE. For SPP =
CC
(26)
instance, at 8:00 h, the proposed IUDI (red) contributed higher Uo of ESC
0.725 than the CVRMSE (blue) with Uo of 0.668, and it is shown that the
where ESC is the cost of energy savings.
proposed IUDI achieved approximately 8% higher Uo than the CVRMSE.
At the same time, for the proposed IUDI, lower PD (i.e., 0.548 kW) was
recorded compared to the CVRMSE (i.e., 0.723 kW), and it is clearly
demonstrated that the proposed IUDI model achieved 13% higher en­
ergy savings than the CVRMSE. Based on the overall performance, the
proposed IUDI achieved 6% improvement of Uo and 27% energy savings
than the CVRMSE throughout working hours. The proposed IUDI
demonstrated great performance in terms of higher visual comfort (Uo)
and energy savings by minimizing the objective functions
simultaneously.

5.4. Technical discussion on the application of the proposed model

Based on the simulation-based results in Section 5.3, the proposed


IUDI with MOPSO can be implemented based on two types of applica­
tions: small-scale and large-scale applications.
In a small-scale application, lighting control can be developed using
a low-cost microcontroller, such as Arduino [26]. In this case, Arduino
Mega 2560 [36] is suitable to be used because the microcontroller can
support all inputs (sensors - analog inputs) and outputs (LED lumi­
naires), which consist of 12 inputs and outputs. The proposed IUDI-­
Fig. 16. Comparison of the best compromise solution by the proposed IUDI and
MOPSO algorithm can be directly transferred to the Arduino Mega 2560
CVRMSE for selected times.
in MATLAB. For sensors, TEMT 6000 ambient light sensor can be

Table 7

Best compromise solutions for dand CVRMSE (FCV).
Result 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00

d1 0.810 0.807 0.879 0.908 0.766 0.819 0.832 0.572 0.587 0.638
d2 0.356 0.143 0.160 0.214 0.142 0.160 0.096 0.154 0.092 0.116
d3 0.678 0.721 0.910 0.773 0.686 0.781 0.783 0.628 0.654 0.693
d4 0.419 0.141 0.045 0.166 0.070 0.076 0.084 0.061 0.099 0.158
d5 0.728 0.494 0.741 0.678 0.613 0.800 0.850 0.668 0.506 0.690
d6 0.303 0.185 0.107 0.161 0.106 0.158 0.085 0.079 0.092 0.098
d7 0.725 0.607 0.851 0.828 0.852 0.874 1.000 0.841 0.917 0.972
d8 0.848 0.974 1.000 1.000 0.930 1.000 1.000 0.954 1.000 0.983
d9 0.765 0.598 0.718 0.748 0.492 0.490 0.534 0.425 0.393 0.381
d10 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000
d11 0.320 0.250 0.374 0.501 0.423 0.481 0.565 0.563 0.751 0.942
d12 0.910 0.978 1.000 1.000 0.964 1.000 1.000 1.000 1.000 1.000

d 6.863 5.899 6.785 6.977 6.043 6.638 6.831 5.944 6.089 6.671
CVRMSE 0.218 0.367 0.405 0.423 0.463 0.435 0.427 0.456 0.426 0.343
PD (kW) 0.723 0.639 0.723 0.737 0.650 0.712 0.727 0.639 0.650 0.703
Uo 0.668 0.627 0.611 0.611 0.618 0.614 0.610 0.615 0.611 0.610

11
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

Fig. 17. Building management system [37].

∑k
ESCt
NPV = (27) Table 9
t − CC
t=1 (1 + r) Detailed calculation of SPP and NPV.
Calculation CVRMSE IUDI
where r is the discount rate.
In this study, a small-scale lighting control system is considered to Capital investment cost
Total system cost (RM) 1349.00
evaluate economic performance. The system consists of three main Total installation cost (RM) 845.00
components: a microcontroller, light sensors, and LED drivers. The de­ Total capital investment cost (RM) 2194.00
tails of the variables for calculating SPP and NPV are presented in Electricity cost
Table 8. Total energy consumption (kWh/year) 1822.66 1332.41
Total electricity cost (RM/year) 665.27 486.33
Based on the variables in Table 8, the detailed calculation of SPP and
Total energy savings (kWh/year) 1318.94 1809.19
NPV is shown in Table 9. The proposed IUDI showed a shorter SPP than Total electricity cost savings (RM/year) 481.41 660.36
the CVRMSE by 28%. Meanwhile, in terms of NPV, the proposed IUDI SPP (year) 4.6 3.3
reported a higher NPV by 33% than the CVRMSE. It can be deduced that NPV (RM) 6658.01 9945.83
the proposed IUDI showed superior performance for both metrics
compared to the CVRMSE.
improve energy performance of a lighting system. In this paper, a new
visual comfort metric is proposed, which corresponds to illuminance
6. Conclusion uniformity (Uo) (i.e., mentioned in the EN 12464-1) called illuminance
uniformity deviation index (IUDI) by considering daylight harvesting.
The main goal of a lighting control strategy is to maximize energy The proposed IUDI is considered as an objective function and minimized
savings and visual comfort of occupants in a building. Daylighting simultaneously with power demand (PD), which is represented by the
should be considered in an optimization-based control strategy to ∑
dimming levels of LED luminaires ( d) (i.e., another objective func­
tion) and optimized using multi-objective particle swarm optimization
Table 8 (MOPSO) algorithm. The proposed model is tested on an office room.
Variables for calculating SPP and NPV. Based on the results, by consideration of individually minimizing the

Variable Value proposed IUDI and d, there are significant improvements for the vi­
Arduino Mega 2560 Microcontroller
sual comfort metric in terms of Uo in the range of 9%–20% and energy
Cost per unit (RM) 179.00 savings in the range of 21%–67% in comparison to the baseline (i.e., no

Installation cost per unit (RM) 200.00 optimization-based control). Meanwhile, by minimizing d and the
Number 1 proposed IUDI simultaneously using MOPSO and then extracting the
TEMT 6000 Ambient Light Sensor
best compromise solution over the Pareto-optimal front using fuzzy-
Cost per unit (RM) 10.00
Installation cost per unit (RM) 10.00 based decision-making approach, Uo and energy efficiency improved
Number 12 by 17% and 58%, respectively, in comparison to the baseline.
Dimmable LED Driver The proposed IUDI is also compared with a variation of root mean
Cost per unit (RM) 30.00 square error (CVRMSE), which is another visual comfort metric model to
Installation cost per unit (RM) 15.00
Number 35
represent Uo in the room. For individual minimization of the objective
Lifetime of system (year) 10 function, the proposed IUDI achieved 7% higher Uo and 15% higher
Electricity tariff (RM/kWh) 0.365 energy savings than the CVRMSE. Meanwhile, for simultaneous bi-
Operating hours (h/year) 2376 objective function minimization, the proposed IUDI improved 6% of
Discount rate (%) 8
Uo and 27% of energy savings compared to the CVRMSE. Apart from

12
K.R. Wagiman et al. Journal of Building Engineering 43 (2021) 102525

improving the technical aspects, the proposed IUDI outperformed the [14] Y. Gao, Y. Cheng, H. Zhang, N. Zou, Dynamic illuminance measurement and
control used for smart lighting with LED, Measurement 139 (2019) 380–386,
CVRMSE for the economic metrics, with the simple payback period
https://doi.org/10.1016/j.measurement.2019.03.003.
(SPP) and net present value (NPV) of 28% and 33%, respectively. [15] S. Ozenc, M. Uzunoglu, O. Guler, Experimental evaluation of the impacts of
Moreover, the proposed IUDI has a practical value to be implemented in considering inherent response characteristics for lighting technologies in building
small- and large-scale applications. In future works, a prototype of the energy modeling, Energy Build. 77 (2014) 432–439, https://doi.org/10.1016/j.
enbuild.2014.03.062.
MOPSO optimization-based control can be developed using [16] L.T. Doulos, A. Tsangrassoulis, P.A. Kontaxis, A. Kontadakis, F.V. Topalis,
microcontroller-based (e.g., Arduino) and implemented in the office Harvesting daylight with LED or T5 fluorescent lamps? The role of dimming,
room to compare the performance between simulation-based and Energy Build. 140 (2017) 336–347, https://doi.org/10.1016/j.
enbuild.2017.02.013.
experimental studies in terms of energy performance and Uo. [17] K.R. Wagiman, M.N. Abdullah, M.Y. Hassan, N.H. Mohammad Radzi, A.H. Abu
Bakar, T.C. Kwang, Lighting system control techniques in commercial buildings:
Author contribution current trends and future directions, J. Build. Eng. 31 (2020) 101342, https://doi.
org/10.1016/j.jobe.2020.101342.
[18] M. Rossi, A. Pandharipande, D. Caicedo, L. Schenato, A. Cenedese, Personal
Khairul Rijal Wagiman: Investigation, Methodology, Software, lighting control with occupancy and daylight adaptation, Energy Build. 105 (2015)
Formal Analysis, Writing – Review & Editing. Mohd Noor Abdullah: 263–272, https://doi.org/10.1016/j.enbuild.2015.07.059.
[19] F. Tan, D. Caicedo, A. Pandharipande, M. Zuniga, Sensor-driven, human-in-the-
Conceptualization, Validation, Resources, Writing – Review & Editing, loop lighting control, Light. Respir. Technol. (2017) 1–21, https://doi.org/
Funding acquisition. Mohammad Yusri Hassan: Supervision, Writing – 10.1177/1477153517693887.
Review & Editing. Nur Hanis Mohammad Radzi: Supervision, Re­ [20] A. Pandharipande, D. Caicedo, Adaptive illumination rendering in LED lighting
systems, IEEE Trans. Syst. Man Cybern. Syst. 43 (2013) 1052–1062, https://doi.
sources, Project administration, Writing – Review & Editing.
org/10.1109/TSMCA.2012.2231859.
[21] D. Caicedo, A. Pandharipande, F.M.J. Willems, Daylight-adaptive lighting control
Declaration of competing interest using light sensor calibration prior-information, Energy Build. 73 (2014) 105–114,
https://doi.org/10.1016/j.enbuild.2014.01.022.
[22] A. Pandharipande, D. Caicedo, Smart indoor lighting systems with luminaire-based
The authors declare that they have no known competing financial sensing: a review of lighting control approaches, Energy Build. 104 (2015)
interests or personal relationships that could have appeared to influence 369–377, https://doi.org/10.1016/j.enbuild.2015.07.035.
the work reported in this paper. [23] D. Caicedo, A. Pandharipande, Daylight and occupancy adaptive lighting control
system: an iterative optimization approach, Light. Res. Technol. 48 (2015)
661–675, https://doi.org/10.1177/1477153515587148.
Acknowledgment [24] D. Caicedo, A. Pandharipande, Sensor-Driven lighting control with illumination
and dimming constraints, IEEE Sens. J. 15 (2015) 5169–5176, https://doi.org/
10.1109/JSEN.2015.2436338.
This work was funded under the Fundamental Research Grant [25] C. Yin, S. Dadras, X. Huang, J. Mei, H. Malek, Y. Cheng, Energy-saving control
Scheme (FRGS/1/2018/TK07/UTHM/02/3) by the Ministry of Higher strategy for lighting system based on multivariate extremum seeking with Newton
Education Malaysia and partially sponsored by Universiti Tun Hussein algorithm, Energy Convers. Manag. 142 (2017) 504–522, https://doi.org/
10.1016/j.enconman.2017.03.072.
Onn Malaysia (UTHM). [26] C. Yin, X. Huang, S. Dadras, Y. Cheng, J. Cao, H. Malek, J. Mei, Design of optimal
lighting control strategy based on multi-variable fractional-order extremum
References seeking method, Inf. Sci. 465 (2018) 38–60, https://doi.org/10.1016/j.
ins.2018.06.059.
[27] W. Si, H. Ogai, T. Li, K. Hirai, A novel energy saving system for office lighting
[1] P.H. Shaikh, N.B.M. Nor, P. Nallagownden, I. Elamvazuthi, T. Ibrahim, A review on
control by using RBFNN and PSO, in: IEEE 2013 Tencon, Spring, 2013,
optimized control systems for building energy and comfort management of smart
pp. 347–351, https://doi.org/10.1109/TENCONSpring.2013.6584469.
sustainable buildings, Renew. Sustain. Energy Rev. 34 (2014) 409–429, https://
[28] W. Si, H. Ogai, K. Hirai, H. Takahashi, M. Ogawa, An improved PSO method for
doi.org/10.1016/j.rser.2014.03.027.
energy saving system of office lighting, SICE Annu. Conf. (2011) 1533–1536, 2011.
[2] EIA, Commercial Buildings Energy Consumption Survey, 2012. https://www.eia.
[29] L.A. Mendes, R.Z. Freire, L. dos S. Coelho, A.S. Moraes, Minimizing computational
gov/consumption/commercial/reports/2012/lighting/. accessed December 22,
cost and energy demand of building lighting systems: a real time experiment using
2019.
a modified competition over resources algorithm, Energy Build. 139 (2017)
[3] The World Bank, Fossil Fuel Energy Consumption, World Bank, 2015. https://data.
108–123, https://doi.org/10.1016/j.enbuild.2016.12.072.
worldbank.org/indicator/eg.use.comm.fo.zs. accessed September 21, 2019.
[30] E.-N.D. Madias, P.A. Kontaxis, F.V. Topalis, Application of multi-objective genetic
[4] I. Nadji Maachi, A. Mokhtari, M.E.-A. Slimani, The natural lighting for energy
algorithms to interior lighting optimization, Energy Build. 125 (2016) 66–74,
saving and visual comfort in collective housing: a case study in the Algerian
https://doi.org/10.1016/j.enbuild.2016.04.078.
building context, J. Build. Eng. 24 (2019) 100760, https://doi.org/10.1016/j.
[31] N. Delgarm, B. Sajadi, F. Kowsary, S. Delgarm, Multi-objective optimization of the
jobe.2019.100760.
building energy performance: a simulation-based approach by means of particle
[5] K.R. Wagiman, M.N. Abdullah, M.Y. Hassan, N.H.M. Radzi, A new optimal light
swarm optimization (PSO), Appl. Energy 170 (2016) 293–303, https://doi.org/
sensor placement method of an indoor lighting control system for improving
10.1016/j.apenergy.2016.02.141.
energy performance and visual comfort, J. Build. Eng. 30 (2020) 101295, https://
[32] R. Yang, L. Wang, Multi-objective optimization for decision-making of energy and
doi.org/10.1016/j.jobe.2020.101295.
comfort management in building automation and control, Sustain. Cities Soc. 2
[6] Q.J. Kwong, Light level, visual comfort and lighting energy savings potential in a
(2012) 1–7, https://doi.org/10.1016/j.scs.2011.09.001.
green-certified high-rise building, J. Build. Eng. 29 (2020) 101198, https://doi.
[33] J. Kennedy, R. Eberhart, Particle swarm optimization, neural networks, in:
org/10.1016/j.jobe.2020.101198.
Proceedings, IEEE Int. Conf., 4, 1995, pp. 1942–1948, https://doi.org/10.1109/
[7] R.M. Ahmad, R.M. Reffat, A comparative study of various daylighting systems in
ICNN.1995.488968, 1995.
office buildings for improving energy efficiency in Egypt, J. Build. Eng. 18 (2018)
[34] Y. Shi, R. Eberhart, A modified particle swarm optimizer, in: 1998 IEEE Int. Conf.
360–376, https://doi.org/10.1016/j.jobe.2018.04.002.
Evol. Comput. Proceedings. IEEE World Congr. Comput. Intell. (Cat.
[8] A. Seyedolhosseini, N. Masoumi, M. Modarressi, N. Karimian, Daylight adaptive
No.98TH8360), 1998, pp. 69–73, https://doi.org/10.1109/ICEC.1998.699146.
smart indoor lighting control method using artificial neural networks, J. Build.
[35] M. Modiri-Delshad, N.A. Rahim, Multi-objective backtracking search algorithm for
Eng. 29 (2020) 101141, https://doi.org/10.1016/j.jobe.2019.101141.
economic emission dispatch problem, Appl. Soft Comput. 40 (2016) 479–494,
[9] A. Al Touma, D. Ouahrani, Quantifying savings in spaces energy demands and CO2
https://doi.org/10.1016/j.asoc.2015.11.020.
emissions by shading and lighting controls in the Arabian Gulf, J. Build. Eng. 18
[36] Arduino, Arduino MEGA 2560, 2017. https://www.arduino.cc/en/Guide/Arduin
(2018) 429–437, https://doi.org/10.1016/j.jobe.2018.04.005.
oMega2560. accessed January 2, 2020.
[10] S. Carlucci, F. Causone, F. De Rosa, L. Pagliano, A review of indices for assessing
[37] T. Park, S. Hong, Experimental case study of a BACnet-based lighting control
visual comfort with a view to their use in optimization processes to support
system, IEEE Trans. Autom. Sci. Eng. 6 (2009) 322–333, https://doi.org/10.1109/
building integrated design, Renew. Sustain. Energy Rev. 47 (2015), https://doi.
TASE.2008.2008148.
org/10.1016/j.rser.2015.03.062.
[38] J.N. Swisher, G. de M. Jannuzzi, R.Y. Redlinger, Tools and Methods for Integrated
[11] European Committee for Standardization, European Standard EN 12464-1: Light
Resource Planning: Improving Energy Efficiency and Protecting the Environment,
and Lighting - Lighting of Work Places - Part 1: Indoor Work Places, 2011.
UNEP Collaborating Centre on Energy and Environment, Risø National Laboratory,
[12] T. Kruisselbrink, R. Dangol, A. Rosemann, Photometric measurements of lighting
Denmark, 1997.
quality: an overview, Build. Environ. 138 (2018) 42–52, https://doi.org/10.1016/
[39] J. Di Stefano, Energy efficiency and the environment: the potential for energy
j.buildenv.2018.04.028.
efficient lighting to save energy and reduce carbon dioxide emissions at Melbourne
[13] S. Borile, A. Pandharipande, D. Caicedo, L. Schenato, A. Cenedese, A Data-Driven
University, Australia, Energy 25 (2000) 823–839, https://doi.org/10.1016/S0360-
Daylight Estimation Approach to Lighting Control, IEEE Access, 2017, p. 1, https://
5442(00)00015-3.
doi.org/10.1109/ACCESS.2017.2679807.

13

You might also like