Energy & Buildings 236 (2021) 110787

Contents lists available at ScienceDirect

Energy & Buildings

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

Calibration of energy simulation using optimization for buildings

with dynamic shading systems
Niraj Kunwar a,1,⇑, Kristen S. Cetin b, Ulrike Passe c
a
Department of Mechanical Engineering, Iowa State University, Ames, IA, USA
b
Department of Civil and Environmental Engineering, Michigan State University, East Lansing, MI, USA
c
College of Design, Iowa State University, Ames, IA, USA

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

Article history: Physics-based energy simulation methods are widely used for the prediction of building energy con-
Received 14 June 2020 sumption and daylighting in commercial buildings. Such simulations are performed to evaluate the
Revised 14 December 2020 design of new buildings or to predict the benefits of various retrofits to the building’s performance.
Accepted 23 January 2021
However, there are uncertainties in the results from such simulations, with higher uncertainty if the
Available online 2 February 2021
model used is not validated empirically. This study focuses on the co-simulation of daylight and energy
modeling to evaluate the energy performance impacts of dynamic shading. The Generalized Pattern
Keywords:
Search algorithm is used to minimize the root mean square error between measured and simulated data
EnergyPlus
Model calibration
to perform the calibration of the baseline energy model while adjusting the building envelope material
Daylighting properties. Next, the results using three different modeling methods for windows and shading devices
Dynamic shading in EnergyPlus were compared to measured data. The results show that the use of the simple layered
Optimization model to represent the dynamic shading system performed better than other two models, when com-
Window pared to the experimental data. They also indicate that part of the difference between the model results
and experimental data originates from sky model calculations and ground reflectance inputs. From the
results it is also observed that using different modeling methods for the window and shading device
can impact the coincidental energy consumption results from cooling and lighting by up to 20%.
Ó 2021 Elsevier B.V. All rights reserved.

1. Introduction
The increasing complexity of building systems and growth of
high performance computing capabilities has increased the use of
computer simulation programs for the prediction of building
energy consumption [1]. Despite the increase in number and capa-
bilities of building energy modeling and simulation software, the
accuracy of prediction of such software is an ongoing point of con-
cern in the built environment sector. In the case of retrofitting
buildings, hybrid models utilizing measured data integrated with
energy simulation methods can be used to help address this con-
cern [2]. This method of model creation is facilitated by the avail-
ability of substantial amounts of data from smart sensors, building
automation systems, and/or building energy management systems.
Currently, the validation of building energy consumption mod-
els is generally performed by comparing the results from simula-
tion models with whole-building or system-level utility data
⇑ Corresponding author at: 813 Bissell Rd, Ames, IA 50011, USA.
E-mail address: nkunwar@iastate.edu (N. Kunwar).
1 both building professionals and consumers who might be wary
Present address: 1 Bethel Valley Rd, Oak Ridge, TN, 37830.
[3,4]. While, comparison with such data may provide decent
results when averaged over a period of time, the heat transfer
dynamics in a building and its systems may not be captured in
the models with utility data-based calibration. Capturing the
building dynamics correctly is important for better prediction of
building energy consumption and thermal comfort, particularly
when model predictions must be extrapolated outside of the initial
bounds of outdoor weather conditions upon which the model was
built. Moreover, short-term building dynamics and their impact on
system operations is of interest as demand side management
(DSM) techniques that use HVAC (heating, ventilation and air con-
ditioning) and other building system components in grid load bal-
ancing and peak shaving as well as occupant comfort. These factors
will be more important moving forward, as buildings become more
integral, dynamic players in grid operations, in response to the use
of more variable renewable energy sources such as wind and solar,
compared to the conventional, base load producing fossil fuel-fired
power plants.
Assessing the accuracy of building energy model predictions is
also needed to provide further evidence of model effectiveness to

https://doi.org/10.1016/j.enbuild.2021.110787
0378-7788/Ó 2021 Elsevier B.V. All rights reserved.
N. Kunwar, K.S. Cetin and U. Passe
of building energy simulation software prediction effectiveness.
Various efforts have been completed to calibrate building energy
models, as well as to address uncertainties [5,6]. Uncertainty and
sensitivity analysis of building energy simulation has been per-
formed considering various parameters, including U-values,
weather variables, infiltration rates, cooling and heating setpoints,
among others [5]. Heo et al. performed Bayesian calibration of nor-
mative building energy models and estimated associated uncer-
tainties using probabilistic predictions of retrofit performance
[6]. A study by Crawley considered the influence of weather data
on the energy demand of an office building [7]. For modeling retro-
fits of existing buildings, one method used to improve prediction
accuracy is the use of hybrid models utilizing measured data inte-
grated with computer simulation [2]. This method of model cre-
ation is facilitated by utilizing significant building performance
and energy data from various sources, such as smart sensors, Build-
ing Energy Management (BEM) and/or Building Automation Sys-
tem (BAS). Another method includes multi-zone calibration using
optimization based on a genetic algorithm [8]. In the study, a
hybrid model is created, where the space conditioning is turned
off during the period of calibration or uncertainty analysis (per-
formed during a free-floating period). During this time, the simu-
lated energy should be zero for the model to be calibrated. The
simultaneous calibration and parameter ranking method for build-
ing energy models found that the cooling plant COP is the most
important factor for model calibration using monthly electricity
data [9]. The calibration of a variable refrigerant flow system mod-
eled in EnergyPlus [10] was performed for an occupancy emulated
small office building [11]. Building energy calibration for a test
building in Belgium was performed by adjusting to the optimal val-
ues for internal gains, ventilation and envelope construction mate-
rial for a short period of monitoring [12]. Calibration has also been
performed utilizing occupancy and plug-load schedules derived
from electricity use data which improved the model accuracy com-
pared to the result obtained using default schedules [13]. In [14],
first Morris sensitivity analysis was used for selection of calibration
parameter, then the calibration was performed based on hourly
heat consumption data. While calibrating the total energy use is
beneficial, it is also important to analyze models of different com-
ponents of building energy simulation software for validating their
accuracy. For example, empirical validation of solar gain through
glazing unit for different energy simulation software [15] and solar
irradiance on inclined surfaces for building energy simulation [16]
was performed. This paper focuses on the models available for
window and shading devices to evaluate their performance against
the measured data using total energy demand as the metric for
evaluation.
In total, it is estimated that 34% of heating and cooling energy in
U.S. commercial buildings is due to the presence of windows [17].
As such, significant research is directed towards the improvement
of the efficiency of windows and shading systems. Experimental
studies have shown that shading systems integrated with lighting
controls have a potential of more than 30% energy consumption
savings in commercial office spaces [18,19]. Simulation studies
have also shown that solar shades can reduce cooling energy sav-
ings by up to 50% [20,21]. For these systems, daylight and visual
comfort prediction is also important in predicting the impact of
different design strategies on buildings. However, there are cur-
rently significant limitations in the abilities of energy simulation
software packages in predicting lighting and daylight parameters
without coupling them with daylight simulation software. For
accurate prediction of daylight, daylight simulation software such
as Radiance [22] and DAYSIM [23] can be used, the results of which
can be integrated into energy simulation software to evaluate the
holistic impact of different daylighting systems on both daylight-
ing and energy consumption[24]. Different studies have focused
2
Energy & Buildings 236 (2021) 110787
on thermal characterization of different types of shading devices
[25–27]. While some studies have addressed the validation of sim-
ulation results against measured data for daylighting using
dynamic shading, the validation of results from energy models is
lacking [28]. In addition, the analysis of the performance of differ-
ent models of solar shades available in widely used energy simula-
tion software such as EnergyPlus [10] is limited. This is because
most of the studies use a single model for solar shades and have
not compare the difference in performance for different modelling
methods of shading device.
This paper addresses the existing gap through the development
and validation of daylight and building energy model of an office
environment with dynamic roller shades and lighting, using con-
trolled laboratory experimental data. The methods include the ini-
tial development of the model, then the optimization was
performed to change the surface material properties used in the
energy model for the baseline case (no shading device) to calibrate
this model against the measured data. Next, shading models are
added to the baseline model together with daylighting-based light-
ing control and a comparison is made between three different
modeling methods for roller shades available in EnergyPlus. A
comparison of daylight modeling with measured data for indoor
illuminance, lighting energy consumption and shading device state
is also made with a DAYSIM based daylight simulation model. In
the results section, to assess the difference between the measured
data and simulation model results, a two-step approach is taken
for both the daylight and energy model. First, the baseline model
is compared with measured data for both models, then in the sec-
ond step, the performance of the model using the shading device is
compared. Finally, the potential sources of differences between the
simulation model and field data are discussed. This research con-
cludes with discussion, conclusions and future work. The results
of this work aids researchers, architects and engineers in choosing
amongst different modeling methods for windows and shading
attachments available in EnergyPlus. The results also show that,
choosing one modeling method versus another can have a signifi-
cant impact on the predicted building energy consumption using
the energy simulation model.
2. Methodology
This section discusses the development of the energy simula-
tion and daylight models for a perimeter office space, the descrip-
tion of which is divided into two subsections. Energy simulation is
performed using the whole building energy simulation engine
EnergyPlus v9.1. Daylight simulation is performed using the day-
light simulation software packages DIVA4-RHINO v4.1 [29] and
DAYSIM v3.0. While EnergyPlus has some built-in daylight simula-
tion capabilities, as well as some capabilities to model shading
devices in different states, these are limited. Custom features to
address the complexities of the shading system and its controls
were required to be added in the simulation in order to attain
the desired output which paralleled the full-scale testing setup
and associated results. The details of full-scale testing which was
conducted for a period of approximately 7 months are briefly dis-
cussed in the following sub-section, and can be found in more
detail in [18]. First, in EnergyPlus, a vertical illuminance sensor
cannot be used. Hence, if a control strategy is based all or in part
on vertical illuminance measurements, as is the case in this effort,
then the control algorithm cannot be directly implemented in
EnergyPlus alone. Second, in EnergyPlus, open loop control of a
shading device is possible. However closed loop control that is con-
trolled based on feedback from interior illuminance, or other mea-
sured value is not currently possible. Since several of the control
strategies utilized in this project were based in part on the vertical
N. Kunwar, K.S. Cetin and U. Passe
illuminance values measured in the interior of the room, in these
cases, DAYSIM was used to control the shading device based on
feedback from vertical illuminance. Here, the vertical illuminance
sensor was used to control the shading device since this illumi-
nance value could directly be used to calculate glare using simpli-
fied daylight glare probability [30]. Finally, output from DAYSIM
was used as input in EnergyPlus to determine the shade status
and lighting dimming.
2.1. Laboratory testing data and test rooms
The laboratory testing data collected and used for this research
includes energy consumption and daylighting data completed and
discussed in detail in [18]. The test facility was located in Ankeny,
IA which lies in ASHREA climate zone 5A. Six identical test rooms
were used during the full-scale testing, including two parallel room
each in the East, South and West orientations. The test room had a
floor area of approximately 26 m2 and ceiling height of 2.52 m with
an exterior window-to-wall ratio of 48%. The images of one of the
test rooms is shown if Appendix A1. Two different types of glazing
were utilized, including double clear glazing in the East test rooms,
double low-e glazing in the West test rooms, and both low-e and
clear glazing in South test rooms (one type in each test room).
The rational of this set-up during the experimental testing is also
discussed in detail in [18]. In this work, a single test room with
exterior windows facing the East orientation is modeled for cali-
bration of the energy model since all the six test rooms were iden-
tical in their geometry and material construction. The single
calibrated room was used to represent results from all the six test
rooms in three different orientations from the full-scale testing
[18] which is further discussed in the section below.
2.2. Shading devices and control strategies
Daylight and energy simulations were performed for two differ-
ent type of roller shades, hereafter named RS1 and RS2, each using
two different control strategies, CS1 and CS2. The experimental
testing of these shading devices and control strategies summarized
below and the resulting impacts on daylighting and energy savings
are discussed in [18]. In this study, the control strategies used dur-
ing experimental testing are modeled and simulated, as the exper-
imental data serves as a basis of validation and calibration of the
models. The properties of the shading devices used are shown in
Table 1.
The two control strategies CS1 and CS2 used [18], are briefly
summarized here. CS1 is based on solar penetration depth and
the illuminance level at the vertical illuminance (VI) sensor located
3 m from the exterior window. The roller shades was modulated to
maintain an illuminance level at the vertical illuminance sensor of
1830 lx ± 110 lx (dead-band). The roller shades was deployed to a
minimum height of the work plane protection height [31] when
the solar irradiation measured at external vertical surface was
greater than 150 W/m2. The shades deployed using work plane
protection height calculated based on solar angles and location of
the work place prevented direct sunlight from reaching the work
plane. CS2 is based on solar penetration depth and the illuminance
level at work plane illuminance (WPI) sensor located 1 m from the
exterior window. The logic of CS2 is similar to CS1; the main differ-
ence being for CS2, the WPI sensor at 1 m from window was used
instead of the VI sensor 3 m from window. All the WPI sensors
were used to evaluate the daylighting in the room and VI sensor
was used to evaluate visual comfort during the experimental test-
ing. As discussed above, for each control strategies only one of
these interior sensors were used for controlling the shading device.
3
Energy & Buildings 236 (2021) 110787
2.3. Daylight modeling
2.3.1. Baseline model (no shading or lighting controls)
The model for daylighting was created using the DIVA-4-RHINO
plugin, which uses DAYSIM as a simulation engine. For each room,
a daylight model was created using the sensor setup shown in
Fig. 1. In the figure, the grid for the lighting control sensor and
the shading control sensor represents the location where the work
plane illuminance sensors were located during laboratory testing.
The sensors for shading control in Fig. 1 is shown for CS2, while
for CS1, VI sensor was used for shading control as discussed in sec-
tion 2.2. The geometry of the Test Rooms was modeled after the
actual geometry of the Test Rooms, then the material properties
(Table 2) for the simulation were assigned. The surface reflectance
properties were assigned to opaque components, while for the
window and the roller shades, visible transmittance was used.
The roller shades were modeled as a translucent material. The
Radiance ‘‘trans material” definition was used for modeling the
roller shades which has capability of specifying diffuse and specu-
lar transmission through a material. It is to be noted that, the roller
shades model discussed is not used for the baseline model but is
only used in model with dynamic shading dicussed in section 2.3.2.
Before the addition of shading controls, a comparison was made
between the measured and simulated illuminance of the three
work plane sensors and one vertical illuminance sensor for the
baseline case without any shading device. The difference between
the measured and the simulated illuminance are summarized in
term of Mean Bias Error (MBE) and Coefficient of Variation of the
Root Mean Squared Error (CV-RMSE) in Table 3. MBE provides a
measure of the bias between two time series datasets, and CV-
RMSE calculates the variation or randomness between the field
data and the model results. These were calculated following the
definitions in ASHRAE Guideline 14 [32]. For MBE, the simulated
data was subtracted from the measured data and the error is rela-
tive to the measured data in terms of percent MBE. MBE essentially
provides the bias between the two time series datasets and CV-
RMSE calculates the variation or randomness between the field
data and the model. These values represent a comparison of mea-
sured and simulated data over a period of one month (April 2017)
using hourly average values for illuminance. Although, there is no
existing guideline for calibration of daylight model, considering the
highly volatile nature of illuminance in the test rooms compared to
other indoor variables like room temperature the model seems to
represent the measured data with a good level of accuracy. In a
previous study on the validation of DAYSIM results, the MBE was
within ± 20% and CV-RMSE was below 32% [33]. In another study
focused on experimental validation of daylight simulation tool
have MBE of up to 39% and CV-RMSE up to 49% [34]. In the latter
study, the measurement was only considered if measured outside
facade illuminance was greater than 5000 lx for the error calcula-
tions. In this study, it was found that the MBE was within ± 20%
and CV-RMSE was below 40% for all the cases. Further the compar-
ison of the measured and simulated illuminance in the South ori-
entation is provided in Fig. 2 as an example of how they compare
to each other.
2.3.2. Model with dynamic shading
For the application of the roller shades control strategies for
daylight simulation, 10 different states of the roller shades were
developed, representing 10% increments of shade height. This
was modeled such that the height of the shading device was chan-
ged based on the feedback from the vertical illuminance sensor at
3 m from the window in the case of CS1, and based on feedback
from the work plane illuminance sensor at 1 m from the window
in case of CS2. The setup of the sensor and application of roller
shades control is shown in Fig. 1.
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Table 1
Properties of the shading devices (roller shades) modeled and tested experimentally.

Shading device Openness factor Visible transmittance Solar transmittance Solar reflectance Fabric color
Roller Shades 1 (RS1) 0.01 0.01 0.01 0.04 Charcoal
Roller Shades 2 (RS2) 0.03 0.12 0.17 0.64 Oyster

during testing, the estimated vertical irradiation on each facade

can be obtained from the results of the energy simulation software.

2.4. Co-simulation method

The model created in DIVA4-Rhino was used to run a simulation

using the daylight simulation software DAYSIM which has capability
to run simulations at a 1-minute timestep, as compared to the 1-
hour interval of DIVA4-Rhino. Using a 1-minute timestep is particu-
larly useful if detailed results are desired for daylight simulation. A
1-minute output is also desirable for use as input into EnergyPlus,
which is also able to run whole building energy simulation in 1-
minute time steps. Thus, the output of shade height from DAYSIM-
based daylight simulation was used as input into energy simulation,
Fig. 1. Daylight model setup for simulated perimeter office space.
given the significant limitations of EnergyPlus in terms of modeling
shading control and daylighting. The comparison of shade height
Table 2
from the experimental data and simulation for the three orienta-
Material properties utilized for modeling. tions are shown in Table 4. Comparison of lighting energy and shade
height for the East orientation using RS2 as the shading device and
Surface Surface reflectance / transmittance (%)
CS1 as the control strategy is shown as an example in Fig. 3.
Ceiling 70
Floor 20
Wall 50 2.5. Energy model
Furniture 50
Clear glazing 80
2.5.1. Baseline model
Low-e glazing 65
Outside Ground 20 The energy modeling was performed using EnergyPlus using the
Roller Shade 1 1 input parameters listed in Table 5. Each test room was modeled as
Roller Shade 2 12 a conditioned zone while the plenum above was modeled as
unconditioned space. The interior walls of each of the six test
rooms and the plenum above was modeled as adiabatic since the
adjacent zone to each test room in the actual building was main-
For lighting control, the artificial interior lighting was set to be tained at a similar temperature to the test room. In addition to
dimmed to maintain a minimum illuminance of 500 lx at a floor the parameters listed in this table, properties of the windows
height of 0.76 m, at work plane illuminance sensor 2.5 m from and shading devices were assigned for their respective test periods
the window, following the full-scale test setup. Next, the output in each orientation. The model utilized an ideal air system to meet
from the both simulations were combined using a custom script the air conditioning requirements of the room. This was compared
in Python to selection the appropriate shade height. An external to the measured cooling load calculated using supply air flow rate
irradiation sensor is not possible to include in DAYSIM 3.0, which and the temperature difference between the supply and room air,
was used as the daylight simulation software. As such, the full- as described in [8]. Lighting power was measured using a Direct
scale experimental method used for control using two sensors, Digital Control (DDC) system. To calibrate the energy model, the
including one external irradiation sensor and an interior illumi- heating ventilation and air conditioning (HVAC) and lighting
nance sensor, cannot be identically modeled in the chosen daylight energy consumption from the measured data and simulated data
simulation software. Therefore, for the application of the utilized for the baseline rooms (no shading or lighting controls) was com-
control algorithm, either measured vertical irradiation, or vertical pared for each orientation. The model parameters in the energy
irradiation at the exterior facade obtained from EnergyPlus output model (material properties) were adjusted to minimize the mean
can be used to obtain the fraction of window occluded by the shad- squared error between the measured and simulated data using
ing device. For this study, the measured external vertical irradia- GenOpt v3.1 [35]. GenOpt is an optimization program that is
tion was used. However, if this information was not measured designed to minimize the cost function evaluated by an external

Table 3
Comparison of measured and simulated illuminance values for the baseline case (no shading or lighting controls) over a period of one month (April 2017).

Orientation WPI (1 m) WPI (2.5 m) VI (3 m) WPI (4 m)

MBE CV-RMSE MBE CV-RSME MBE CV-RMSE MBE CV-RMSE
East 3 25 16 15 15 32 14 14
South 1 28 13 12 7 13 13 9
West 18 30 6 17 15 39 1 17

Note: WPI = work plane illuminance sensor; VI = vertical illuminance sensor; distance shown in meters is distance from the exterior window; MBE = mean bias error; CV-
RMSE = Coefficient of Variation of the Root Mean Squared Error.

4
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Fig. 2. Comparison of illuminance in the South baseline test rooms from April 1st-7th, 2017 at (a) WPI 1 m (b) WPI 2.5 m (c) VI 3 m (d) WPI 4 m.

Table 4
Comparison of simulation result and measured data for lighting energy use (kWh) and shade height (m).

MBE CV-RMSE
Shade Type RS1 RS2 RS1 RS2
Control Strategy CS1 CS2 CS1 CS2 CS1 CS2 CS1 CS2
Lighting energy (kWh) East 1.7 1.6 1.6 0.5 8.0 11.1 4.8 7.3
South 4.0 1.0 5.0 5.7 9.8 7.9 11.5 12.5
West 0.3 3.4 1.4 1.7 8.5 17.2 7.1 9.5
Shade height (m) East 16.4 2.6 18.5 7.8 38.8 19.2 39.3 21.6
South 43.7 5.8 18.8 21.1 73.4 16.6 80.0 56.2
West 6.3 11.8 1.1 18.5 73.7 41.0 42.5 39.0

Note: RS1 = roller shade 1; RS2 = roller shade 2 (see Table 1 for properties); CS1 = control strategy 1; CS2 = control strategy 2; MBE = Mean bias error; CV-RMSE = coefficient of
variance of root mean square error.

Fig. 3. Time series comparison of (a) lighting fraction (%) and (b) roller shades closed status (%), from daylight simulation and field data for a period of 6 days.

5
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Table 5 2.5.2. Model with dynamic shading

Input parameters for energy model development. In EnergyPlus, roller shades can be modeled using three differ-
Parameters Value ent methods, including (a) the simple window model, (b) the com-
Weather data (EPW format) Custom weather file created from on- plex fenestration model, and (c) the equivalent layer model [36].
site weather station data for the period Details on these models are described in [36], some details of
of testing which are discussed in this section. Using the built-in functions,
Occupancy schedule and internal 6:30 am to 9:30 pm EnergyPlus can only control roller shades to be in either the ‘‘on”
loads from office equipment
(fully ON)
(fully closed) or ‘‘off” (fully open) position. Because of this limita-
Zone floor area 25.9 m2 tion, roller shades cannot be directly controlled at intermediate
Zone volume 65.3 m3 heights in EnergyPlus using currently integrated techniques. To
Ideal load air system Ideal load air system with reheat with overcome this limitation, the window was divided into 10 horizon-
rated maximum flow rate of 0.47 m3/s
tal sections, each accounting for 10% of the actual window size.
per room and rated minimum flow rate
of 0.037 m3/s per room Then, using the shade height as an annual schedule, the shading
device was closed/opened to the respective height for each win-
dow segment. For example, if the shading height was considered
to be 10% closed, the first, top section of the window’s shading
simulation program such as EnergyPlus. It was developed for prob-
device state was set to ‘‘closed” and other nine sections’ shading
lems where the cost function is computationally expensive [35].
device states were set to ‘‘open”. Similarly, if the shading height
The Hooke-Jeeves algorithm was used for the Generalized Pattern
was set to 20%, the first and second top sections of the window
Search (GPS) to identify the optimal solution for the model calibra-
shading device state were set to ‘‘closed” and the rest were set to
tion. Using the updated model, energy simulation was then per-
‘‘open”. Here, the shade height resulting from the daylight simula-
formed for the roller shades control. The geometry of the energy
tion can be used. However, in this study, we used the measured
model (left) and the step-by-step process of model calibration fol-
shade height such that any error originating from the differences
lowed herein (right) are shown in Fig. 4. Several of the important
in measured and simulated shade height is not introduced into
parameters that were used during the building energy simulation
the model. The description of the three methods mentioned above
are listed in Table 5. The change in energy consumption resulting
are as follows:
from the optimization is shown in Fig. 5.
This optimization was carried out for the test room in the East
1. Simple layered window: To model the window, this model uses
orientation, then the room was rotated 90° and 180° for the simu-
different layers of glazing and gaps filled with different air or
lation in the South and West orientation respectively. The glazing
gas mixtures. For the clear glazing, the glazing layers were both
properties were changed for the test rooms with low-e glazing to
clear and 3 mm in thickness, and the air gap was 6.3 mm thick.
correspond to the field-tested conditions.
The calculated center of glass visible transmittance (VT) was
Adjustments were made to the material in the construction of
0.81, U-value was 3.12 W/m2-K and the solar heat gain coeffi-
the test rooms which include the changes to the material proper-
cient (SHGC) was 0.76 when evaluated in WINDOW [37]. For
ties in four different locations, including Mat1 (ceiling), Mat2 (in-
low-e glazing, the glazing layers were low-e 3 mm and clear
ternal wall), Mat3 (internal wall), Mat4 (external wall) and Mat5
3 mm and the gap was 11 mm with a 90% argon-10% air mix-
(floor). The original properties and optimized properties of these
ture. The low-e double glazing has center of glass VT of 0.65,
layers are shown in Fig. 6. Here, the materials are named Mat1,
U-value of 1.36 W/m2-K and solar heat gain coefficient of
Mat2, etc. because after optimization the material properties might
0.24. In the remainder of this work, this model will be
not represent the actual material properties of the original mate-
addressed as the ‘‘Simple model”.
rial. The properties of the original materials that were used for
2. Equivalent layer model: This model uses direct and diffuse split
optimization are included in Appendix A2. In the figure, the origi-
of the incident solar irradiation to represent the shading device.
nal and optimized properties for thickness (T), thermal conductiv-
The determination of off-normal solar optical properties using
ity (k), specific heat capacity (c) and density (d) is provided.
this model is based on [26]. The model is not intended for mate-
The changes in the material properties changed the heat capac-
rials with high values of openness and should be limited to a
ity of the test rooms, calculated as sum of the product of mass and
maximum openness fraction of 0.20.
specific heat capacity of each of the materials, from 19.5 Million J/k
3. Complex Fenestration System (CFS) model: Here, bi-directional
to 25.3 Million J/k. Measured energy consumption data was used
scattering distribution function (BSDF) data exported from
for model calibration to adjust building material property values.
WINDOW v7.6 [37] was used to model the shading device.
As such, the resulting model may be considered a hybrid model
Two BSDF files, one without a shading device and other with
which uses a physics-based simulation engine to develop and
a shading device were imported in EnergyPlus. The input used
run the simulations, and information from measured data to
for calculating the BSDF for the glazing was the same construc-
inform parameter adjustments. Such adjustments are made while
tion that was used for creating the simple layered window. One
preserving the physical relationship and associated impact of the
of the shades (RS2), already had a BSDF file available in the
window and shading device on model in terms of heat gains and
WINDOW database while for another (RS1) the BSDF file
losses for which the measured or manufacturer provided values
needed to define the shading layer. This was created using the
were used throughout the simulation. Similar approaches of using
genBSDF tool in Radiance [22]. For both the equivalent layer
optimization techniques for model calibration through the adjust-
model and complex fenestration model, an energy management
ment of building material properties have been used in [8,12]. The
system (EMS) in EnergyPlus was used to change the construc-
resulting value of the material properties obtained using this
tion of the window material to represent two different scenar-
method also consider the impact and uncertainties from the ther-
ios, including with and without the shading device. The
mal mass of furniture and equipment, thermal bridging and air
construction of different sections of the windows among the
infiltration, among other properties. Therefore, the utilized values
10 horizontal sections were changed based on the shade height
in the simulation deviate from the actual physical value since they
schedule created based on measured data.
may account for such uncertainties.

6
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Fig. 4. Energy model geometry of a single perimeter office space test room (left) and calibration procedure (right).

Fig. 5. Comparison of simulation model before and after optimization with measured data.

Fig. 6. Material properties changes before and after optimization (Note: Ori = Original; Opt = Optimized).

7
N. Kunwar, K.S. Cetin and U. Passe
Table 6
Comparison of measured and simulated energy consumption for the baseline case (no shading or lighting controls) in each orientation for a period of one month (July 1 to 30,
2017).
Measured (kWh) Simulated (kWh)
CFS Equivalent
East 867.3 922.6 795.0
South 631.1 724.1 615.6
West 586.7 624.0 620.3
3. Results
3.1. Energy model
3.1.1. Baseline energy model comparison with measured data
The comparison of total energy consumption between the mea-
sured and simulated data at hourly intervals for the baseline (no
shading or lighting controls) test case over a period of one month
(July 1st to 31st, 2017) is provided in Table 6. This table provides
the total value of measured and simulated energy and MBE and
CV-RMSE. Table 6 shows that the for the rooms with clear glazing,
i.e. in East and South orientation, the equivalent layer model com-
putes lower energy consumption compared to the CFS and Simple
model. The MBE is within ± 15% for all models, and the CV-RMSE is
below 40%. Graphs from July 1st to 15th, for hourly cooling energy
are shown in Fig. 7, the graph is only provided for 15 days for the
better visualization. Fig. 7 shows that for the low-e glazing in the
West orientation, the performance of three different models is sim-
ilar and there is a slight over-prediction of energy consumption
during the morning hours and under-prediction during the evening
hours as compared to the measured data.
3.1.2. Model with dynamic shading
The results of the energy simulation that used the three differ-
ent models for the roller shades are compared to the laboratory
measured data. The comparison between the measured and the
simulated data for the Test Rooms with both dynamic shading
and lighting dimming controls are summarized in Table 7. The
table includes the MBE and CV-RMSE values, as well as the mea-
sured and simulated energy use results.The energy use values in
Table 7 include coincidental energy consumption from both cool-
ing and lighting applications.
As shown in the Table 7, the MBE is higher for CFS and Equiva-
lent model compared to the Simple model. The MBE value is posi-
tive for almost all the test cases while using the CFS and Equivalent
model, while this is either positive or negative for different test
cases of the Simple model. This shows that overall measured
energy use is higher compared to simulation model using CFS
and the Equivalent models. In addition, the energy consumption
while using the Simple model is higher compared to the other
two models. This difference in the energy consumption originates
from the cooling energy demand since the same lighting energy
demand obtained from daylight simulation was used for all mod-
els. The MBE values in the table show that for the Simple model,
the MBE is within ± 15% while the value is higher in other models.
The comparison of the three models to the field data is further
shown in Fig. 8. The figure provides the comparison of cooling
energy consumption from the measured data with the three mod-
els when using both the shading devices for CS1. The figure pro-
vides the results for all three orientations. In Fig. 8, it can be seen
that both the Equivalent layer model and the CFS model calculate
lower energy consumption compared to the Simple model. These
results are interesting since both the CFS and the Equivalent layer
model should perform better in comparison to the Simple model.
Both CFS and Equivalent layer models use more detailed input
Energy & Buildings 236 (2021) 110787
MBE CV-RMSE
Simple CFS Equivalent Simple CFS Equivalent Simple
911.8 6.4 8.3 5.1 39.8 38.4 39.5
692.5 14.7 2.5 13.0 34.9 25.1 33.1
630.8 6.4 5.7 7.5 36.0 36.1 36.5
compared to the Simple model and consider the impact of the solar
profile angle on the transmission and reflection properties of the
shading device. However, for the majority of the cases, the Simple
model seems to capture the dynamics of the energy consumption
of the measured data better compared to the other two models.
In a past study, it was found that different models of shading
devices showed RMSE for illuminance that differed by up to
600 lx [38]. However, in the case of energy impact of shade models
to the best knowledge of authors, the difference between the all
three models and measured data is not reported before. From the
results obtained in this study, it can be concluded that more exper-
iments on shade that is adjusted at different heights of window/s
should be performed. The comparison of the experimental results
to different models is needed to determine the model that has
superior performance to reflect reality with more confidence.
There could be various reasons for the difference between the
measured and simulated data, including uncertainty inherent in
the model or measured data, sky radiance distribution calculation
using Perez sky model [39] used by EnergyPlus, and non-uniform
ground reflectance values. While, it is not within the scope of this
paper to identify all the potential sources of error, one is discussed
here. It was found that there was difference in incident external
vertical irradiation that was measured during the test and calcu-
lated using the energy simulation model as shown in Fig. 9. In
the figure, we can see that in the South orientation the simulated
external vertical irradiation (EVI) is higher than the measured val-
ues. This corresponds to the cooling energy demand in Fig. 8,
where the peak load during the afternoon hour in the South orien-
tation is higher in the simulated data compared to the measured
data. Similarly, the measured external vertical irradiation is higher
in the West orientation for both RS1 and RS2 compared to the sim-
ulated value. This also causes the cooling demand during the late
afternoon hours in the West orientation to be lower during the
simulation compared to the measured data. Since measured global
horizontal irradiation, and direct normal irradiation was used to
create the measured data, the potential source of error for the
EVI should be the combination of the errors originating from the
sky model and ground reflectance values.
The comparison between the measured and the simulated data
in this section provided empirical validation for the Simple Model
which shows that MBE and CV-RMSE for the Simple Model with
roller shades is comparable to the case without any shading device.
However, a larger difference was seen with the Equivalent layer
and the CFS model, with lower energy consumption from these
two models compared to the Simple model and the measured data.
The energy consumption using these models are lower during the
period when direct sunlight hits the facade. For example, this
occurs during morning hours in the east and the evening hours
in the west orientation. The difference in these models and their
accuracy should be further investigated to identify which of the
three models is more accurate when used for different types of
shading devices. Since there have been no previous studies that
conducted field testing of roller shades for their energy impact,
the results of this study are among the first to show that one
modeled in EnergyPlus performs better than the other two when
8
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Fig. 7. Measured and simulated hourly cooling energy demand for the (a) East, (b) South, and (c) West baseline Test Room for July 1–15, 2017.

Table 7
Comparison of measured and simulated energy consumption data for case with dynamic roller shades and lighting dimming controls.

Shade Control MBE CV-RMSE Measured (kWh) Simulated (kWh)

CFS Eqv Simple CFS Eqv Simple CFS Eqv Simple
East RS1 CS1 40.0 26.7 5.6 61.3 41.9 36.9 299.2 179.6 219.3 316.0
CS2 34.8 23.9 8.6 57.1 39.8 41.9 390.2 254.5 297.0 423.6
RS2 CS1 33.2 34.7 13.4 49.7 50.2 36.3 117.8 78.7 77.0 102.0
CS2 23.0 26.3 0.5 36.6 38.6 29.1 124.6 95.9 91.9 125.2
South RS1 CS1 22.6 18.3 9.3 45.4 43.3 43.8 124.7 96.5 101.9 113.1
CS2 20.4 15.7 5.8 37.8 35.3 34.9 127.6 101.6 107.6 120.2
RS2 CS1 5.2 1.3 7.3 24.8 24.3 29.1 121.9 115.6 120.3 130.8
CS2 2.6 1.3 9.9 25.4 25.5 30.7 152.3 148.3 154.2 167.4
West RS1 CS1 15.6 10.7 1.2 45.6 43.0 39.8 139.9 118.1 125.0 138.2
CS2 20.6 14.2 1.9 51.1 47.3 43.6 153.5 121.8 131.7 150.5
RS2 CS1 16.9 10.2 1.8 40.3 38.7 36.7 228.2 189.5 204.9 232.2
CS2 18.2 11.8 1.1 39.8 37.7 37.4 227.8 186.5 201.0 225.4

Note: RS = roller shades; CS = control strategy; MBE = mean bias error; RMSE = root mean squared error; CFS = complex fenestration system; Eqv = Equivalent.

compared to measured data. Future studies can be conducted to
assess the sensitivity of different modeling approach to properties
such as U-value and SHGC of the window system when used with
different types of shading devices.
4. Conclusions
Daylighting and energy models were created for six laboratory
office spaces using dynamic shading strategies. First, a comparison
of performance from the daylight model was completed to
measured data for indoor illuminance, shade state and lighting
energy consumption. Then, the model calibration was performed
using the measured data and a generalized pattern search opti-
mization for the baseline case for the energy model. The method
applied for model calibration using cooling energy and lighting
energy demand can be used to predict the load of an existing build-
ing before assessing the impact of HVAC equipment/efficiency.
Models calibrated with this approach can be used to determine

9
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Fig. 8. Comparison of measured data and three different shading models in the East, West and South orientation using CS1 for RS1 and RS2.

Fig. 9. Comparison of measured data and simulation of External Vertical Illuminance (EVI) data for the East, West and South orientation.

potential optimal control strategies for HVAC equipment to meet
the cooling load or reduce the peak demand. After performing
the baseline calibration, the shade model was added to the cali-
brated model. Then, a comparison of the different models available
was conducted against the measured data. The comparison results
did not show noticeable differences in terms of MBE and CV-RMSE
across the considered shading devices and control strategies. How-
ever, the comparison showed that the results from the different
energy models can vary by more than 20% while using different
available modeling methods. It was also observed that the simple

10
N. Kunwar, K.S. Cetin and U. Passe Energy & Buildings 236 (2021) 110787

Fig. A1. Photos of test room exterior (left), interior (right).

Table A2 Resources, Writing - review & editing, Supervision. Ulrike Passe:

Details of contruction of the test room surfaces. Conceptualization, Funding acquisition, Investigation, Writing -
Construction Layers (Outside Type of Material in Thermal resistance review & editing, Supervision.
to Inside) EnergyPlus (m2-K/W)
Ceiling Ceramic/clay Material NA
tiles (Mat1) Declaration of Competing Interest
Interior Wall Gypsum Material NA
Plastering The authors declare that they have no known competing finan-
(Mat2)
cial interests or personal relationships that could have appeared
Fiber Glass Material NA
(Mat3) to influence the work reported in this paper.
Gypsum Material NA
Plastering
(Mat2) Acknowledgements
Exterior W1_RVAL_1 Material:NoMass 0.077
Wall W1_RVAL_2 Material:NoMass 1.6526
W1_RVAL_3 Material:NoMass 0.16896
This material is based upon work funded by ASHRAE under the
Gypsum Board Material NA grant 1710-RP, supported by Iowa Energy Center and Pella Corpo-
(Mat4) ration and the work supported by National Science Foundation
Floor Cast Concrete Material NA under Grant No. (1762592/2013093). Any opinions, findings, and
(Mat5)
conclusions or recommendations expressed in this material are
Roof Roof_RVAL_1 Material:NoMass 0.0176
Roof_RVAL_2 Material:NoMass 0.07744 those of the author(s) and do not necessarily reflect the views of
Roof_RVAL_3 Material:NoMass 4.42112 ASHRAE, the Iowa Energy Center, Pella Corporation and National
Roof_RVAL_4 Material:NoMass 0.03872 Science Foundation. We would like to thank Dr. Xiaohui Zhou
Roof_RVAL_5 Material:NoMass 0.15312 and Dr. Yunhua Li for their support and feedback during the vari-
Roof_RVAL_6 Material:NoMass 0.03872
ous stages of the project.

layered model for the window performed better in comparison to Appendix

the other two models in EnergyPlus when the model was com-
pared to the measured data for the case with the shading device.
This suggests that further investigation of the window and shade
model in EnergyPlus might be needed so that each of the model
can provide accurate results for energy consumption.
A limitation of this study is that it considers the calculated
room-level cooling loads to be equal to cooling energy consump-
tion, when in reality, the type of HVAC system and associated effi-
ciency would impact the final cooling energy consumption
measured. However, given the room level modeling, it was not pos-
sible to measure energy consumption associated with individual
rooms’ loads. When considering total energy consumption reduc-
tion across both HVAC and lighting, the final percent and total
energy use calculated would be impacted. In future studies, a sys-
tem level analysis could be performed to determine the actual
cooling energy demand, which also would show the potential of
integrated controls for shading, lighting and HVAC”.
CRediT authorship contribution statement
Niraj Kunwar: Data curation, Methodology, Formal analysis,
Writing - original draft, Software. Kristen S. Cetin: Conceptualiza-
tion, Funding acquisition, Investigation, Project administration,
11
A1. Details of the window size
The area of the exterior wall is 4.47 m (width)  3.05 m (height)
while the surface area of the windows in each room is approxi-
mately 4.27 m (width)  1.52 m (height). Therefore, the WWR is
approximately 48%. Fig. A1 provides an exterior view and interior
view of the test room showing the size of the window relative to
the exterior wall.”
A2. Construction of test room
The details of construction of test room surfaces are provided in
Table A2. For calibration using optimization only the material
properties of the type ‘‘Material” in Table A2 which is defined by
material properties thermal conductivity, specific heat capacity,
density and thickness in EnergyPlus was modified. These proper-
ties have already been discussed in the main text. The other type
of material ‘‘Material:NoMass” was defined only using a thermal
resistance value as shown in Table A2 was and was not modified
during the calibration process. The value of the thermal resistance
for different layer of material defined as ‘‘Material:NoMass” is pro-
vided in the table. The Mat1 to Mat5 that were used for optimiza-
tion are also provided to show the original properties of the
materials that were optimized.
N. Kunwar, K.S. Cetin and U. Passe
Appendix A. Supplementary data
Supplementary data to this article can be found online at
https://doi.org/10.1016/j.enbuild.2021.110787.
References
[1] T. Hong, S.K. Chou, T.Y. Bong, Building simulation: an overview of
developments and information sources, Build. Environ. 35 (4) (2000) 347–
361, https://doi.org/10.1016/S0360-1323(99)00023-2.
[2] D. Coakley, P. Raftery, M. Keane, A review of methods to match building energy
simulation models to measured data, Renew. Sustain. Energy Rev. 37 (2014)
123–141, https://doi.org/10.1016/j.rser.2014.05.007.
[3] Y. Pan, Z. Huang, G. Wu, Calibrated building energy simulation and its
application in a high-rise commercial building in Shanghai, Energy Build. 39
(6) (2007) 651–657, https://doi.org/10.1016/j.enbuild.2006.09.013.
[4] A. Pedrini, F.S. Westphal, R. Lamberts, A methodology for building energy
modelling and calibration in warm climates, Build. Environ. 37 (8-9) (2002)
903–912, https://doi.org/10.1016/S0360-1323(02)00051-3.
[5] G. Calleja Rodríguez, A. Carrillo Andrés, F. Domínguez Muñoz, J.M. Cejudo
López, Y. Zhang, Uncertainties and sensitivity analysis in building energy
simulation using macroparameters, Energy Build. 67 (2013) 79–87, https://doi.
org/10.1016/j.enbuild.2013.08.009.
[6] Y. Heo, R. Choudhary, G.A. Augenbroe, Calibration of building energy models
for retrofit analysis under uncertainty, Energy Build. 47 (2012) 550–560,
https://doi.org/10.1016/j.enbuild.2011.12.029.
[7] D.B. Crawley, Which weather data should you use for energy simulations of
commercial buildings?, ASHRAE Trans. 104 (1998) 498–515.
[8] C. Fernández Bandera, G. Ramos Ruiz, Towards a new generation of building
envelope calibration, Energies 10 (12) (2017) 2102, https://doi.org/10.3390/
en10122102.
[9] J. Yuan, V. Nian, B. Su, Q. Meng, A simultaneous calibration and parameter
ranking method for building energy models, Appl. Energy. 206 (2017) 657–
666, https://doi.org/10.1016/j.apenergy.2017.08.220.
[10] US DOE, NREL, EnergyPlus, (2019). https://energyplus.net/.
[11] D. Kim, S.J. Cox, H. Cho, P. Im, Model calibration of a variable refrigerant flow
system with a dedicated outdoor air system: a case study, Energy Build. 158
(2018) 884–896, https://doi.org/10.1016/j.enbuild.2017.10.049.
[12] V. Monetti, E. Davin, E. Fabrizio, P. André, M. Filippi, Calibration of building
energy simulation models based on optimization: a case study, Energy
Procedia 78 (2015) 2971–2976, https://doi.org/10.1016/j.egypro.2015.11.693.
[13] Y.S. Kim, M. Heidarinejad, M. Dahlhausen, J. Srebric, Building energy model
calibration with schedules derived from electricity use data, Appl. Energy 190
(2017) 997–1007, https://doi.org/10.1016/j.apenergy.2016.12.167.
[14] W. Li, Z. Tian, Y. Lu, F. Fu, Stepwise calibration for residential building thermal
performance model using hourly heat consumption data, Energy Build. 181
(2018) 10–25, https://doi.org/10.1016/j.enbuild.2018.10.001.
[15] P. Loutzenhiser, H. Manz, P. Strachan, C. Felsmann, T. Frank, G. Maxwell, P.
Oelhafen, An empirical validation of modeling solar gains through a glazing
unit using building energy simulation programs, HVAC R Res. 12 (4) (2006)
1097–1116.
[16] P.G. Loutzenhiser, H. Manz, C. Felsmann, P.A. Strachan, T. Frank, G.M. Maxwell,
Empirical validation of models to compute solar irradiance on inclined
surfaces for building energy simulation, Sol. Energy. 81 (2) (2007) 254–267,
https://doi.org/10.1016/j.solener.2006.03.009.
[17] J. Apte, D. Arasteh, Window-Related Energy Consumption in the US Residential
and Commercial Building Stock, Lawrence Berkeley Natl. Lab. LBNL-60146
(2008).
12
Energy & Buildings 236 (2021) 110787
[18] N. Kunwar, K.S. Cetin, U. Passe, X. Zhou, Y. Li, Full-scale experimental testing of
integrated dynamically-operated roller shades and lighting in perimeter office
spaces, Sol. Energy 186 (2019) 17–28, https://doi.org/10.1016/
j.solener.2019.04.069.
[19] N. Kunwar, K.S. Cetin, U. Passe, X. Zhou, Y. Li, Energy savings and daylighting
evaluation of dynamic venetian blinds and lighting through full-scale
experimental testing, Energy 197 (2020) 117190, https://doi.org/10.1016/j.
energy.2020.117190.
[20] A. Tzempelikos, A.K. Athienitis, The impact of shading design and control on
building cooling and lighting demand, Sol. Energy 81 (3) (2007) 369–382,
https://doi.org/10.1016/j.solener.2006.06.015.
[21] N. Kunwar, M. Bhandari, A Comprehensive Analysis of Energy and Daylighting
Impact of Window Shading Systems and Control Strategies on Commercial
Buildings in the United States 13 (9) (2020) 2401, https://doi.org/10.3390/
en13092401.
[22] LBNL, Radiance, (2019). https://www.radiance-online.org/.
[23] M.S.D. Lab, DAYSIM, (2019). http://daysim.ning.com/.
[24] Y. Sun, Y. Wu, R. Wilson, A review of thermal and optical characterisation of
complex window systems and their building performance prediction, Appl.
Energy 222 (2018) 729–747, https://doi.org/10.1016/j.apenergy.2018.03.144.
[25] J.L. Wright, Calculating center-glass performance indices of glazing systems
with shading devices, ASHRAE Trans. 114 (2008) 199–209, https://doi.org/
10.1017/CBO9781107415324.004.
[26] N.A. Kotey, J.L. Wright, M.R. Collins, Determing off-normal solar optical
properties of roller blinds, ASHRAE Trans. 115 (2009) 10.
[27] N. Kotey, M. Collins, J.L. Wright, T. Tiang, A simplified method for calculating
the effective solar optical properties of a venetian blind layer for building
energy simulation, J. Sol. Energy Eng. 131 (2009) 9, https://doi.org/10.1115/
1.3090822.
[28] H. Shen, A. Tzempelikos, Daylight-linked synchronized shading operation
using simplified model-based control, Energy Build. 145 (2017) 200–212,
https://doi.org/10.1016/j.enbuild.2017.04.021.
[29] L.L.C. Solemma, DIVA-FOR-RHINO V4 (2019). http://diva4rhino.com/.
[30] J. Wienold, Dynamic simulation of blind control strategies for visual comfort
and energy balance analysis, IBPSA 2007, Int Build. Perform. Simul. Assoc. 2007
(2007) 1197–1204, https://doi.org/10.1016/S0378-7788(01)00058-5.
[31] A. Tzempelikos, H. Shen, Comparative control strategies for roller shades with
respect to daylighting and energy performance, Build. Environ. 67 (2013) 179–
192, https://doi.org/10.1016/j.buildenv.2013.05.016.
[32] American Society of Heating Refrigeration and Air-Conditioning Engineers,
Measurement of Energy, Demand and Water Savings, 2014.
[33] C.F. Reinhart, O. Walkenhorst, Validation of dynamic RADIANCE-based
daylight simulations for a test office with external blinds, Energy Build. 33
(7) (2001) 683–697, https://doi.org/10.1016/S0378-7788(01)00058-5.
[34] C. Reinhart, P.F. Breton, Experimental validation of 3ds MaxÒ design 2009 and
Daysim 3.0, IBPSA 2009, Int. Build. Perform. Simul. Assoc. (2009, 2009,) 1514–
1521.
[35] LBNL, GenOpt, (2019). https://simulationresearch.lbl.gov/GO/.
[36] Engineering Reference, 2019. https://energyplus.net/sites/
all/modules/custom/nrel_custom/pdfs/pdfs_v9.3.0/EngineeringReference.pdf.
[37] LBNL, WINDOW, 2019. (n.d.). https://windows.lbl.gov/software/window.
[38] Y.C. Chan, A. Tzempelikos, Solar optical properties of roller shades: modeling
approaches, measured results and impact on energy use and visual comfort,
3rd High Perform. Build. Conf. Purdue. (2014) 1–14.
[39] R. Perez, R. Seals, J. Michalsky, All-weather model for sky luminance
distribution-preliminary configuration and validation, Sol. Energy 50 (3)
(1993) 235–245, https://doi.org/10.1016/0038-092X(93)90017-I.