You are on page 1of 13

Energy & Buildings 253 (2021) 111529

Contents lists available at ScienceDirect

Energy & Buildings


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

An integrated optimization and sensitivity analysis approach to support


the life cycle energy trade-off in building design
Jani Mukkavaara a,⇑, Farshid Shadram b
a
Industrialized and Sustainable Construction, Department of Civil, Environmental and Natural Resources Engineering, Luleå University of Technology, Sweden
b
Civil Engineering and Built Environment, Department of Civil and Industrial Engineering, Uppsala University, Sweden

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

Article history: The building design process plays a central role in efforts to implement energy-efficient practices.
Received 17 May 2021 However, unilateral design choices based solely on reducing operational energy use can significantly
Revised 15 September 2021 increase a building’s embodied energy and life cycle energy use as there is a trade-off between embodied
Accepted 28 September 2021
and operational energy. To support such trade-off problems, multi-objective optimization represents a
Available online 30 September 2021
useful approach that produces a set of optimal solutions from where a solution can then be selected
and progressed within the design process. Selecting one solution from the set of optimal solutions can
Keywords:
however be a challenging task as each solution has the potential to be chosen as the optimum.
Embodied energy
Energy efficiency measures
Therefore, the purpose of this study was to explore how solutions from a multi-objective optimization
Life cycle energy approach can be analyzed further to provide information to decision-makers when selecting the optimal
Multi-objective optimization design solution. An approach is proposed where the integration of post-optimization sensitivity analysis
Operational energy into a multi-objective optimization approach aims to support decision-makers in analyzing the optimal
Sensitivity analysis solutions provided by the optimization process. The applicability of approach is demonstrated using a
case of a multifamily apartment building located in Sweden, where the aforementioned trade-off is
explored for a set of energy efficiency measures. Thereby, a diverse range of optimal solutions that could
result in up to 4520 GJ life cycle energy (LCE) savings relative to the case building’s initial design was ini-
tially identified using the multi-objective optimization. These solutions were then subjected to a sensi-
tivity analysis where the results indicated that in general the lowest and highest sensitivity in terms
of LCE use belonged to the insulation thicknesses in roof and walls, respectively. Furthermore, the thick-
ness of exterior floor insulation yielded the greatest variation in the sensitivity. The findings of case study
indicate that the post-optimization sensitivity analysis can add valuable information that complements
the results obtained using a multi-objective optimization approach. Consequently, it can support
decision-making on how to progress with the design in terms of what design parameters have a negligi-
ble or significant impact on the objectives when they are varied, thus facilitating prioritization.
Ó 2021 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).

1. Introduction national and international goals/targets regarding energy use and


related CO2 emissions in buildings [4–6]. Many European countries
1.1. Life cycle energy in buildings have developed requirements and certification systems to pursue a
reduction in buildings’ energy use. In Sweden, the national board of
The building industry is one of the largest users of energy and housing, building, and planning has set requirements for improv-
material resources worldwide [1]. In the European Union, buildings ing a building’s annual energy demand (i.e. energy use for annual
account for almost 40% of final energy use [2]. There is thus a grow- heating, cooling, and hot water demand). Hence, a comprehensive
ing awareness within the industry concerning buildings’ energy understanding and novel approaches are sought after to achieve
use and their impact on the environment [3]. Consequently, sus- these targets and fulfill the requirements on a building’s energy
tainable or green building design interests have brought ambitious use [7]. However, recent studies also emphasize the significance
of embodied energy (defined in this paper as the energy used for
the production of materials and components) and considering
⇑ Corresponding author at: Luleå University of Technology, Laboratorievägen 14, buildings’ energy use from a life cycle perspective [8–10]. The
971 87 Luleå, Sweden. background to this comes from the use of energy efficiency
E-mail address: jani.mukkavaara@ltu.se (J. Mukkavaara).

https://doi.org/10.1016/j.enbuild.2021.111529
0378-7788/Ó 2021 The Author(s). Published by Elsevier B.V.
This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

Nomenclature

Ed,s annual energy use for an energy supply source [MJ]


EEe embodied energy of a construction element [MJ/m2] Greek letters
EEm embodied energy of a material [MJ/m3] m* absolute mean elementary effect [-]
EF1-M exterior floor insulation material [-]
EF1-T exterior floor insulation thickness [m] Abbreviations
ENAe enclosing area of a construction element [m2] BIM building information modeling
EW1-M exterior wall (concrete) insulation material [-]
CAD computer-aided design
EW2-M exterior wall (wood studs) insulation material [-] EE embodied energy
EW1-T exterior wall (concrete) insulation thickness [m] IDF input data format
EW2-T exterior wall (wood studs) insulation thickness [m] LCE life cycle energy
l building lifespan [yr]
NSGA-II non-dominated sorting genetic algorithm
Ps primary energy factor for an energy supply source [-] OE operational energy
RF1-M roof insulation material [-] SRRC standardized rank regression coefficient
RF1-T roof insulation thickness [m]
tm thickness of a material [m]

measures to reduce a building’s energy use during its operational that typical objectives for building optimization studies were
phase (or operational energy), e.g. use of thick and/or high- related to environmental impact, initial investment and opera-
performance insulation materials, energy-efficient windows, and tional cost, as well as thermal comfort. Additionally, they con-
efficient heat ventilation air conditioning (HVAC) systems. cluded that construction materials, building geometry and
Although these energy efficiency measures can minimize the orientation, and HVAC system design were typical design parame-
energy use during a building’s operational phase, their use can ters subjected to optimization.
increase the embodied energy to an extent more than the offset- In particular, multi-objective optimization is suggested as a
ting reductions obtained in the operational energy, which is unfa- promising approach as there often exist multiple conflicting objec-
vorable from a life cycle energy perspective [8–10]. A review of 90 tives in real-world building design problems [14,16]. As such,
life cycle energy assessment studies of buildings by Chastas et al. multi-objective optimization can, in many cases, be more relevant
[8] indicates an increased share of embodied energy in the transi- than a single-objective approach. This is also reflected in research
tion from conventional to low energy and near-zero energy build- efforts, where there has been an increase in applying multi-
ings, which reached up to 50% of a building’s total life cycle energy objective optimization [13,14]. For example, in a recent study, Jalali
use. Similarly, Kneifel et al. [10] showed that increasing a build- et al. [17] used genetic algorithms with parametric modeling to
ing’s operational energy efficiency by adopting energy efficiency optimize the shape of a building and the window-to-wall ratio to
measures corresponds with increased embodied energy flows in maximize both natural light and indoor space. D’Agostino et al.
which near-zero energy performance can result in an over 40% [18] proposed a computational design approach that uses multi-
increase in the building’s embodied energy use. The aforemen- objective optimization for exploring a building’s energy need, day-
tioned studies highlight that a trade-off exists between a building’s light, and construction cost. Rosso et al. [19] used multi-objective
energy use during its operation and its embodied energy, where optimization to minimize annual energy use and cost, investment
the application of energy efficiency measures minimizes the for- cost and CO2 emissions of building retrofit in the Mediterranean
mer’s contribution while increasing the latter’s share. Therefore, climate. A comprehensive multi-objective optimization framework
it is of significance to explore and understand how various energy is developed by Ascione et al. [20] for building energy design in
efficiency measures can affect the embodied/operational energy order to minimize its primary energy use, global cost and CO2
trade-off already during the building design process in order to emissions. Ghaderian and Veysi [21] combined surrogate model
make decisions that favorably minimize energy use from a life with a multi-objective evolutionary algorithm to improve occu-
cycle perspective. pants’ thermal comfort and simultaneously minimize building’s
energy use. Pazouki et al. [22] developed a fuzzy robust multi-
1.2. Multi-objective optimization objective optimization model to find optimal energy retrofit solu-
tion for a university building. Jung et al. [23] used multi-
Trial-and-error methods based on parametric analysis are one objective optimization to find optimal passive design strategy for
of the ways that have traditionally been applied to explore design multi-story residential building in terms of energy, environmental
choices for trade-off problems [11]. However, approaches based on and economic metrics from a life cycle perspective. A multi-
changing one parameter while holding the others constant can objective optimization approach was developed by Ciardiello
neglect potentially important interactive effects, and full paramet- et al. [24] to optimize the shape and envelope in building energy
ric studies can be time-consuming and infeasible [11]. The advent design. Chegari et al. [25] developed a multi-objective optimization
of computational design approaches, which strive to further approach by combining artificial neural networks and metaheuris-
exploit the computational capabilities of modern computers, has tic algorithms to improve building energy performance and indoor
provided a foundation to implement optimization approaches that thermal comfort. A multi-objective optimization approach was
aim to find optimal solutions to trade-off problems in design developed by Acar et al. [26] to minimize overall thermal energy
[12,13]. Due to the potential power of these approaches, there need and initial investment cost of residential buildings in turkey.
has been an increasing trend in the number of building optimiza- In relation to life cycle energy, Shadram & Mukkavaara [27,28]
tion studies, which has demonstrated a great interest from the used multi-objective optimization in two different studies to
research community on the issue [14,15]. In a review of explore the trade-off between embodied and operational energy
optimization of building designs, Machairas et al. [16] concluded with design parameters related to construction materials in the

2
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

building’s envelope and the building’s geometry. Similarly, Abbasi the search space and screen out insignificant inputs that have no
& Noorzai [29] proposed a multi-objective optimization approach or negligible influence on the objectives, a process referred to as
for the exploration of embodied and operational energy use, where factor fixing [37,41]. For example, Eisenhower et al. [31] used sen-
they also integrated considerations regarding renewable energy in sitivity analysis to reduce the number of parameters in their opti-
the optimization process. These studies show how optimization mization approach with building energy simulation meta-models.
approaches could provide valuable insights for decision-makers They found that by using sensitivity analysis they could perform
on potentially optimal solutions and the impact of different mea- optimization over a smaller subset of influential parameters with
sures on conflicting objectives during the design process. results comparable to optimization using all parameters. Similarly,
Even though previous research has provided ample findings that Zajas & Heiselberg [42] used sensitivity analysis in their study of
highlight the potential benefits of applying multi-objective opti- thermal properties concerning window frame geometry. Their
mization, there are still challenges and obstacles ahead. This study found that by conducting sensitivity analysis pre-
includes, for example, reducing the computational cost, considering optimization they could reduce the parameters used in the opti-
the uncertainties of design variables, environmental variables, mod- mization process and, consequently, reduce the computational
els, managing problem definition and scale as well as analysis and time needed. These approaches of applying sensitivity analysis
better understanding of optimization results [13,14]. In this paper, can thus provide valuable efforts in simplifications of models and
the focus is on the challenge of providing a better understanding reductions in computational time if needed, which can be useful
of the results that are provided from applying an optimization in practical applications.
approach, a topic that is argued to become more important as the Besides applying sensitivity analysis pre-optimization, it could
applications of multi-objective optimization for building design also be useful post-optimization to assist decision-making and bet-
grow in scope [13]. This process includes interpreting the optimiza- ter the understanding of how sensitive the obtained optimal solu-
tion results and making decisions on how to progress with the tions (Pareto solutions) are with respect to changes in the design
design during post-optimization. The root of this challenge is parameters. This can be particularly useful as the possibility to
derived from what the output is from applying a multi-objective include all criteria (e.g. constructability, logistics, procurement)
optimization approach. In particular, this is seen in relation to Pareto in an optimization process is limited and thus the obtained optimal
optimization strategies, which are a common way of performing solutions must be further analyzed before progressing with the
multi-objective optimization [13,16]. Within the concept of Pareto design. This process is conceptually similar to that conducted by,
optimality, a set of trade-off optimal solutions (referred to as Pareto for example, Heiselberg et al. [34] that used sensitivity analysis
solutions) is produced. This means that the output from such a to provide input of the significance and relevance of design deci-
multi-objective optimization approach is a set of solutions, instead sions with respect to a building’s performance. Similarly, Østergård
of a single solution. What is noteworthy is that each of these Pareto et al. [37] used sensitivity analysis and building performance sim-
solutions can be considered optimal in the absence of further infor- ulation to guide designers in preventing unwanted performance
mation [30]. Consequently, to progress the design process, decision- changes. However, these were performed without the context of
makers need to analyze and select the most appropriate solution first applying multi-objective optimization. An example of sensi-
from these Pareto solutions, a task that can be challenging [14]. tivity analysis being applied post-optimization can be found in
One approach to supporting such a process is to provide more infor- Tuhus-Dubrow & Krati [11] where they used an optimization
mation on the behavior of the optimal solutions, for example regard- approach to investigate building shape and envelope design with
ing how optimal solutions are affected by variations in the design the objective to minimize energy use. They used a simple sensitiv-
parameters, which can assist decision-making. In line with this, it ity analysis post-optimization to investigate the effects of design
has been argued that integration of multi-objective optimization environment parameters (climate, utility rates, and heating and
with sensitivity analysis increases their potential to be adopted cooling set-points) on their optimal solutions.
when supporting design decisions [31].
1.4. Research aim and scope
1.3. Sensitivity analysis
Although there has been an increasing interest in using multi-
Sensitivity analysis is a concept that concerns measuring the objective optimization and sensitivity analysis to explore building
effect of a given input of a model on its output [32]. One of the performance and sustainable design, there have been limited
motivations behind applying sensitivity analysis in building design efforts to adopt such approaches for exploring the impact of energy
is to increase the understanding of the relationships and relative efficiency measures on the embodied/operational energy trade-off,
importance of different design parameters. This can provide possi- despite its importance from a life cycle energy perspective. Addi-
bilities to improve building performance by carefully considering tionally, and central to this study, sensitivity analysis has previ-
the important design parameters [33]. The results of sensitivity ously been applied mainly pre-optimization to screen out
analyses can thus provide valuable information about what design insignificant parameters that have no or negligible influence on
parameters to focus on as the design progresses and give informa- the objectives in order to minimize the solution space and opti-
tion about less influential design parameters that will only have a mization costs. Sensitivity analysis as a post-optimization process
minor impact on building performance [34]. This can provide to guide decision-makers in analyzing further optimal solutions
decision-makers with the information needed to better understand (i.e. Pareto solutions) has received limited attention, despite its
a design problem and how design parameters interact with the potentially valuable contributions [11]. This is particularly relevant
problem and its objectives. Because of this potential, sensitivity as the task of selecting the most appropriate optimal solution to
analyses have been applied in several previously published studies progress the design with is not necessarily a trivial task, which
on building design (e.g. [35–40]). consequently needs to be supported [14]. To address this gap,
In relation to multi-objective optimization, sensitivity analysis the aim of this study was to propose and test a post-
could be used in several different ways. This can be discussed using optimization sensitivity analysis approach that integrates multi-
the pre- or post-optimization phases (i.e. actions performed before objective optimization with sensitivity analysis to find and analyze
or after an optimization approach, respectively) employed by optimal solutions that consider the embodied/operational energy
Nguyen et al. [14]. Application of sensitivity analysis during a trade-off in order to make design decisions that favorably mini-
pre-optimization phase can facilitate a reduction in the size of mize a building’s energy use from a life cycle perspective.
3
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

An overview of the scope for a building’s life cycle energy use is By combining the information gathered for the product defini-
presented in Fig. 1. The focus of this study was primarily on the tion and design parameters, a parametric model can be defined.
embodied/operational energy trade-off caused by materials and The parametric model represents a way of describing a design
components used in the building envelope. The embodied energy symbolically based on parameters [12] and can be used to generate
is defined to include the energy used for off-site production of solutions through a rule-based process [45]. This can be imple-
building materials and components. The operational energy is mented by capturing relevant parameters, relationships, con-
defined in this study to include the energy use during the build- straints, conditions, attributes, and functions [46]. By altering the
ing’s lifespan to produce hot water and heating the building, as values of the design parameters in the parametric model, different
well as the operational electricity use, but excludes household solutions can be generated; a process that both the optimization
electricity use because of its dependence on the devices present algorithm and sensitivity analysis interact with. Implementation
in the individual households. of the parametric model can, for example, be done using CAD
(computer-aided design) or BIM (building information modeling)
software [47,48]. Herein, the parametric model was implemented
2. Method through a combination of the input data format (IDF) for the
dynamic energy simulation tool EnergyPlus and the Python pro-
2.1. Multi-objective optimization and sensitivity analysis approach gramming language. An IDF file was used to encapsulate the pro-
duct definition and a Python script was used to modify that IDF
The outcome from this study is the proposal of an approach, file with the current values of the design parameters.
which is used to illustrate the general strategy for solving the iden-
tified problem. The approach is based on leveraging a computa-
2.1.2. Evaluate solution
tional design setup, where multi-objective optimization and
When a solution has been generated using the parametric
sensitivity analysis is linked with a parametric model and a series
model through variation of the design parameters it needs to be
of evaluations. The approach was iteratively developed together
evaluated to provide input to the multi-objective optimization
with a prototype, which is used to demonstrate and evaluate the
and sensitivity analysis. The evaluation process, together with
approach with the case building described in Section 2.2. Fig. 2
the generation of solutions using the parametric model, is central
shows an overview of the approach and a more detailed descrip-
to leveraging a computational design approach as it provides feed-
tion of its components follows below.
back that can be used to develop the design [12]. The specific eval-
uations that are needed are dependent on the objectives of the
2.1.1. Product definition, design parameters and parametric model design problem and can include calculations, simulations, analyses,
The initial step of the proposed approach is constituted by set- or other processes deemed relevant [49]. The key here is that the
ting up a product definition, design parameters, and a parametric generation of a solution based on the parametric model and the
model; a concept adapted from [43]. The product definition con- evaluation of that solution is performed automatically for both
tains the static traits that are used to define the overall product. the optimization and sensitivity analysis to function efficiently.
For the case demonstration in this paper, the product definition This is important as hundreds or thousands of solutions need to
consists of a 3D model representing the overall geometry and ther- be generated and evaluated.
mal zones of a building together with attributes corresponding to For the case demonstration in this paper, two different evalua-
the building’s geographic location and construction elements. The tions were conducted: dynamic energy simulation and quantity
design parameters are used to establish the boundaries within take-off. For the dynamic energy simulation, EnergyPlus was used
which possible solutions exist by defining variations in relevant to obtain the annual energy performance of the building and the
aspects of the design. Herein, these correspond to relevant energy operative temperature in different zones. The quantity take-off
efficiency measures that are of interest to the design. Each design was implemented using a Python script and was used to calculate
parameter can be defined either in continuous (a range defined the quantities of each construction element and their constituent
by lower and upper limits) or discrete form (a set of alternatives). material quantities.
A design parameter can affect the physical design (e.g. through
dimensions) or its properties (e.g. thermal properties through 2.1.3. Multi-objective optimization
material selection). Because of the inherent effects of defining To conduct the multi-objective optimization there is a variety of
design parameters on the range of possible solutions, careful con- different algorithms and approaches that can be used [14]. Herein,
sideration should be given to their choices such that they can lead a stochastic population-based genetic algorithm was used. This
to more meaningful results [44]. Material types, material quanti- type of algorithm has been indicated in previous studies to be
ties, and component types related to the building’s envelope are robust at large-scale problems and especially concerning disconti-
used as the design parameters for the case demonstration in this nuities, such as those that may occur in the output due to the use of
paper, which are further detailed in Section 2.2. discrete variables [14,50]. More specifically, the non-dominated

Fig. 1. An overview of the scope of a building’s life cycle energy in this study.

4
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

Fig. 2. Overview of the proposed approach for the integration of multi-objective optimization and sensitivity analysis. Boxes encapsulated by dashed lines indicates case
specific implementations made for the study in this paper.

sorting genetic algorithm (NSGA-II, see [30]) was used, which was construction element u) and tm (m) is the thickness of material m
chosen as it is commonly used for trade-off problems related to in each construction element. The materials’ embodied energy
building design [13]. Additionally, Pareto optimization was used was primarily obtained from the Bath Inventory of Carbon and
as the optimization method to identify optimal solutions from a Energy (ICE) [52], however, for some components such as win-
set of feasible trade-off solutions that are generated by the dows, the embodied energy data was gathered from EPD databases
NSGA-II algorithm. A solution is deemed Pareto optimal when no [53,54]. The Bath ICE was used here since (1) it is an open access
other feasible solutions exist that improve one objective without database and contains data for only building materials compared
worsening at least one other objective. Pareto optimization was with other databases that include material data from other sectors,
used instead of, for example, scalarization where weight factors and (2) there is lack of a Swedish national database that contains
are assigned to each criterion to construct an objective function, the embodied energy data for various building materials.
primarily due to the inherent difficulties in estimating and The second objective function (i.e. the operational energy) was
attributing the weight factors to the design objectives [14]. Addi- estimated using the following equation:
tionally, using Pareto optimization can expose the optimization
  Xy
problem and the optimal solutions to more in-depth analysis f 2 x ¼ ð s¼1 Ed;s Ps Þ  l ð4Þ
where different potentially optimal solutions can be presented
and further analyzed to gain knowledge on the properties of opti- In equation (4), Ed;s (MJ) is the annual energy use for energy sup-
mal solutions [51]. The output from the optimization process is ply source s (where s is an index ranging from zero to the total
thus a set of solutions, herein referred to as Pareto solutions, each number of relevant energy supply sources y). The annual energy
of which could be considered optimal in the absence of further use for different energy supply sources was an output from Energy-
information [30]. Plus (the dynamic energy simulation in the 2nd step). P s is the pri-
For the case demonstration in this paper the multi-objective mary energy factor for the energy supply source s, which was
optimization problem was defined as follows: obtained from the Swedish building code [55]. Finally, l (yr) is
     the lifespan of the building.
min f 1 x ; f 2 x ; x ¼ ½x1 ; x2 ;    ; xn  ð1Þ Each of the trade-off solutions (represented as different values
in the design parameters) is created by the optimization algorithm
where f 1 is the first objective function (i.e. embodied energy), and f 2
and passed through the parametric model to generate a model for
is the second objective function (i.e. operational energy), x is a com- the associated solution. The model is then evaluated using the 2nd
bination of design parameters x1 ; x2 ;    ; xm , and n is the number of step, which provides the input needed for computing the objective
design parameters. functions (as detailed above) and constraints (requirements of
The first objective function (i.e. embodied energy) was esti- Swedish building codes, e.g. the allowed heat transfer coefficient,
mated using the following equations: annual primary energy, and operative temperature in various
  Xj zones). The optimization process is performed until the defined
f1 x ¼ e¼1
EEe  ENAe ð2Þ termination condition is met (e.g. the maximum number of
generations).
Xu
EEe ¼ EEm  t m ð3Þ The multi-objective optimization through the NSGA-II algo-
m¼1
rithm was implemented using the pymoo package for the Python
In equation (2), EEe (MJ/m2) is the embodied energy of construc- programming language (see [56]). In general, Ascione et al. [20]
tion element e (where e is an index ranging from zero to the total indicated that in terms of reliability and convergence of multi-
number of construction elements in the building j), and ENAe (m2) objective genetic algorithm, the most important parameters are
is the enclosing area of the construction element e. In equation (3), maximum number of generations and population size. More
EEm (MJ/m3) is the embodied energy of material m (where m is an specifically, the values for these parameters are suggested to be
index ranging from zero to the total number of materials in each 10–100 as maximum generations and a population size equivalent
5
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

to 2–6 times the number of design parameters [20]. As the case approximation of the total sensitivity of each parameter including
study building (see section 2.2) investigates nine design parame- their interactions with other parameters [37]. This measure can be
ters, the NSGA-II parameters were set up with a termination con- suitable for detecting highly influential parameters and insignifi-
dition of maximum number of generations = 100 and population cant parameters that have no or negligible influence on the design
size = 50. The other NSGA-II related parameters were (based on objectives. Additionally, a sensitivity analysis approach was chosen
[30]) set as crossover probability = 0.9, mutation probability = 1/ where the parameters are given a small interval of variation
n (where n is the number of design parameters), and crowding around a nominal value. The nominal value is here derived from
degree = 20. Additionally, it was set up to be performed using par- the design parameter values of the selected Pareto solutions. The
allelization to decrease its overall run time. small interval of variation provides a range within which each
parameter can vary from its nominal value. For the study in this
2.1.4. Select solutions for further analysis paper an interval of ±0.02 m was chosen for the thickness param-
As noted previously, with a Pareto-based multi-objective opti- eters in the sensitivity analysis. This interval selection was made to
mization the results of the optimization process will be a set of Par- keep the variance in the obtained solutions close to the known
eto solutions. Depending on the optimization problem and its parameter values (i.e. those provided by the multi-objective opti-
definition, this can range from only a few solutions to a substantial mization) [39], thus keeping them situational instead of going for
number of solutions. To progress into the sensitivity analysis a set a broad interval that represents the diversity of the full design
of relevant solutions needs to be selected from the Pareto space. During the study both smaller and larger intervals were
solutions. tested (i.e. ± 0.01 m and ± 0.03 m) with similar results. It is of note
It should be recognized that it is theoretically possible to per- that the interval value of ±0.02 m is inherently relative and depen-
form post-optimization sensitivity analysis on all Pareto solutions. dent on the study context and could be assumed smaller or greater
However, this has two main practical challenges. First, depending depending on the parameters it is attached to.
on the choice of sensitivity analysis method, a qualitative evalua- The process of performing the sensitivity analysis is conducted
tion needs to be made, i.e. the sensitivity analysis only provides in two stages. First, a set of samples are generated that represent
information that a user then manually needs to assess in order to the model inputs. Each of the generated samples is then passed
make decisions on how to progress. If this needs to be done on a through the parametric model to generate a model representing
large number of solutions it can become impractical and poten- that solution, which is then evaluated to extract the model outputs.
tially overwhelming. Additionally, if continuous design parameters In the second stage, the actual sensitivity analysis is conducted,
are used there is a possibility that many Pareto solutions are very where the model inputs and outputs are subjected to analysis.
similar to each other, which will also result in a sensitivity analysis The output is a set of sensitivity indices (including m*) for each
with a similar outcome. Second, if many Pareto solutions are design parameter and objective. The outcome of this process indi-
passed through the sensitivity analysis there is a computational cates the sensitivity of the obtained optimal solution(s) from the
cost that needs to be accounted for, where the feasible amount optimization approach with respect to the variations in the design
of time spent should be considered. parameters. These results can then add valuable information for
There are many ways to reduce the size of the Pareto solution design decisions and progression that complements the results
set, including ranking techniques for multi-criteria decision mak- obtained using the multi-objective optimization approach.
ing [57] and clustering techniques, such as K-means [58]. Herein, Implementation of the sensitivity analysis using the Method of
a simple technique was used which groups Pareto solutions based Morris was done using the sensitivity analysis library (SALib) for
on their discrete design parameter values and chooses a median the Python programming language (see [62]). It was set up with
solution from those groups based on their performance. This tech- 50 trajectories and 4 levels. Additionally, like the multi-objective
nique was deemed a suitable approach as it provided a simple way optimization, it was also set up to be performed using paralleliza-
of both reducing the size of the Pareto solution set while still pro- tion to decrease its overall run time.
viding a diverse set of solutions to further evaluate and analyze in To ensure that the sensitivity analysis approach applied herein
terms of their sensitivity to embodied and operational energy. provided reliable results, assessment of the Standardized Rank
Regression Coefficient (SRRC) [41] was also tested, as it is another
2.1.5. Generate samples and sensitivity analysis a widely used sensitivity analysis approach in the building perfor-
The last step of the approach is to conduct sensitivity analysis mance research context [63–65]. This was done by generating
on the selected Pareto solutions. The purpose of this step is to gain samples using Latin Hypercube Sampling and performing the cal-
a better understanding of what design parameters have the culation of the SRRCs using the R programming language and the
greatest influence on the design objectives and should be given sensitivity package (see [66]). The findings showed that the results
extra consideration when moving forward in the design process of applying the Method of Morris (as suggested herein) and
[34]. Similar to multi-objective optimization, there are multiple assessment of the SRRCs were comparable.
different sensitivity analysis methods, and they are valuable and
appropriate for different purposes [33,38,41,59]. For the case 2.2. Description of the case building and design parameters
demonstration in this paper, the Method of Morris was used, which
is one of the most commonly used screening-based methods for For this study, an apartment building is used as a case to
sensitivity analysis [59]. This method can provide a compromise demonstrate and evaluate the applicability of the proposed
between accuracy and efficiency, compared to, for example, approach. The apartment building is designed based on a standard-
variance-based methods that generally provide better information ized building concept, which is characterized by high levels of pre-
but has a much higher computational cost [39]. engineering that aims to shorten the design process and contribute
The Method of Morris uses the concept of elementary effects to to cost-effective solutions. The apartment building consists of 39
characterize the sensitivity of a model with respect to its input apartments distributed across eight stories, a total heated floor
parameters, where the elementary effects are approximations of area of 2478 m2, and a glazing area of 340 m2. The heating system
the first-order partial derivatives of the model [60]. The method uses hydronic radiators connected to the city’s district heating grid.
is thus based on computing the elementary effects of each input, The building is also equipped with an air-to-air heat exchange ven-
which are then averaged to provide an assessment of the inputs’ tilation system (with 80% efficiency). The apartment building is
overall importance [61]. These values (denoted m*) provides an located in Vindeln, a locality situated in the northern parts of
6
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

Sweden, approximately 545 km (geodesic distance) north of Stock- zones. These represent the intended use of the as-built design as a
holm, Sweden. The climate conditions in Vindeln are subarctic, multifamily residential building consisting of only apartments and
characterized by cold and dark winter days with an annual mean were obtained from the case company as well as the Swedish
outdoor temperature of 3.4 °C. building codes, guidelines, and energy standards [55,69]. With
Data for the case building was collected from drawings and a respect to the thermal zones, each floor was considered as a sepa-
BIM model provided by the case company. This provided input rate zone. The occupancy schedule was defined so that zones are
for the product definition, selection of design parameters, and con- occupied from the afternoon until the morning (i.e. not during
sequently the definition of the parametric model. An illustration working hours), and the HVAC schedule was set to be always on.
and floor plan for the building can be seen in Fig. 3. Lastly, the building’s lifespan was considered to be 50 years.
To set up the parametric model the apartment building’s initial
design was used as a starting point. In total 9 design parameters
were included where eight of them were related to insulation in 3. Results and analysis
the construction elements and one was related to the window
types. Tables 1–4 show the constituent materials of the exterior The results of executing the multi-objective optimization (i.e.
construction elements and the window types used in the building, step 3 in the proposed approach, see Fig. 2) on the case building
together with the design parameters subjected to the optimization are presented in Fig. 4 as the differences in embodied and opera-
and sensitivity analysis. As shown in Table 2, four design parame- tional energy relative to the building’s initial design. This is shown
ters (i.e. material quantities) were defined in a continuous form for all obtained feasible solutions (n = 4920), Pareto solutions
and the rest (i.e. material types and windows) were defined as dis- (n = 420), and the selected solutions (n = 12) later subject to sen-
crete. This use of both continuous and discrete parameters was sitivity analysis (i.e. step 5 in the proposed approach, see Fig. 2).
made in line with previous studies that share similar contexts Each solution identified by the multi-objective optimization pro-
[27,28,49]. However, it should be noted that the material quanti- cess represents a unique combination of the design parameters
ties could also be defined as discrete parameters using set intervals listed in Table 2.
(e.g. 0.01 m), which has the potential to reduce the overall compu- In general, the Pareto solutions yielded a range from a reduction
tational cost and provide more solutions that are more easily trans- in embodied energy of 129 GJ to an increase of 1911 GJ compared
lated into practical solutions. to the building’s initial design. For operational energy, the range is
The insulation materials that were considered included both from a reduction of 5230 GJ to an increase of 106 GJ. The results
traditional ones, such as mineral wool, cellulose, and expanded thus contain solutions with a variety of different configurations
polystyrene (EPS), and high-performance ones, such as polyisocya- that represent a broad range of performance considering both
nurate (PIR). In terms of embodied energy, PIR exhibits a higher embodied and operational energy. Furthermore, these results also
value compared to mineral wool, cellulose, and EPS while provid- indicate that the application of the energy efficiency measures
ing lower thermal conductivity (see, for example, [67,68]). listed in Table 2 for the case building can result in larger reductions
Thereby, the use of traditional insulation materials can require in operational energy compared to increases in embodied energy.
thicker building envelopes to achieve sufficiently low operational The results in Fig. 4 also show the trade-off between embodied
energy standards compared with high-performance ones. Pursuing and operational energy, where reductions in one increase the
very thick building envelopes may not be feasible due to consider- other’s share. This trade-off is less noticeable in the Pareto solution
ations including architectural appearance and space limitations. providing the minimum embodied energy in which the reductions
The range of insulation thicknesses (see Table 2) was thus defined in the embodied energy (129 GJ) and increase in operational
based on practical considerations of appearance and available energy (106 GJ) were similar in scope (see Fig. 4, Min. EE). How-
space in the case building. ever, by pursuing additional energy efficiency measures and striv-
The configurations and profiles that were used for evaluating ing to further decrease the operational energy the results show
the operational energy in the case demonstration are presented that there is a more considerable trade-off in the Pareto solutions.
in Table 5, where the values were commonly used for all the heated This can be seen, for example, regarding the Pareto solutions pro-

Fig. 3. 3D-model illustrating the case building and a 2D-drawing showing the building’s floor plan.

7
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

Table 1
The exterior components and construction elements associated with the as-built design of the apartment building and the associated constituent type or material subject to
optimization.

Component/construction Enclosing area Initial thermal transmittance Initial Initial Design parameter Design parameter
element (m2) (W/(m2 K)) type/material thickness (m) (type/material) (thickness)
Windows (WN1) 336 1.0 Standard triple- N/A WN1-M N/A
glazed
Exterior floor (EF1) 340 0.33 EPS 0.100 EF1-M EF1-T
Exterior wall, concrete 502 0.2 Mineral wool 0.170 EW1-M EW1-T
(EW1)
Exterior wall, wood studs 1068 0.15 Mineral wool 0.080 EW2-M EW2-T
(EW2)
Roof (RF1) 340 0.09 Cellulose 0.450 RF1-M RF1-T

Table 2 Table 5
A list of the design parameters, their ranges or values, and types used in this study. Configurations and profiles used in the dynamic energy simulation in this study.

Design Range/values Type Configuration Value


parameter
Room’s temperature set point (heating) 21 °C 1
WN1-M Standard triple-glazed, passive house Discrete Cooling N/A
window Hot water demand 25 kWh/(m2 yr) 1
EF1-M EPS, PIR Discrete Internal gains from domestic hot water usage 20 % 1
EF1-T 0.06 – 0.3 m Continuous Occupant density 0.033 occupants/
EW1-M Mineral wool, EPS, PIR Discrete m2 1
EW1-T 0.08 – 0.3 m Continuous Occupant presence per day 14 h 1
EW2-M Mineral wool, EPS, PIR Discrete Effect per occupant 80 W 1
EW2-T 0 – 0.25 m Continuous Internal gains from occupants’ heat 100 % 1
RF1-M Cellulose, mineral wool, EPS, PIR Discrete Mechanical ventilation (air flow) 0.35 l/(s m2)
RF1-T 0.1 – 0.6 m Continuous Infiltration rate (constant) 0.6 l/(s m2 surface
area)
Additional energy use and losses (e.g., distribution 10 % of the heating
system losses, plant losses, and thermal bridges) demand 2
Table 3
1
A list of the insulation materials and their properties used for the design parameters Sveby, standards for energy in buildings. Input data for calculation of building’s
in this study. energy use. http://www.sveby.org/wp-content/uploads/2012/10/Sveby_Brukar-
indata_bostader_version_1.0.pdf. Accessed 04/01, 2021.
Material Thermal conductivity (W/ Embodied energy 2
F. Bianchi, A.l. Pisello, G. Baldinelli, F. Infrared thermography assessment of
(m K)) (MJ/m3)
thermal bridges in building envelope: Experimental validation in a test room setup.
Cellulose 0.042 106 Sustainability 6(10) (2014) 7107–7120, https:///https://doi.org/
Mineral wool 0.039 830 10.3390/su6107107.
Expanded polystyrene 0.035 2658
(EPS)
Polyisocyanurate (PIR) 0.028 3063 of the first and necessary measures to launch a reduction in life
cycle energy use for the building case.
Beyond the window type (WN1-M) parameter, the configura-
Table 4 tion of the Pareto solutions generally indicated that further reduc-
A list of the components and their properties used for the design parameters in this
tions in operational energy entail increasing the thicknesses of the
study.
insulation materials (i.e. EF1-T, EW1-T, EW2-T, RF1-T) and moving
Component Thermal transmittance (W/ Embodied energy (MJ/ towards implementing higher-performing materials (e.g. PIR) for
(m2 K)) m2)
the insulation (i.e. EF1-M, EW1-M, EW2-M, RF1-M). The solution
Standard triple- 1.0 1462 with the lowest noted life cycle energy (according to the scope pre-
glazed
sented in Fig. 1; see Fig. 4, Min. LCE) increased the insulation thick-
Passive house 0.8 1585
window nesses for the exterior walls (EW1-T, EW2-T) and roof (RF1-T) to
their maximum allowed value (see Table 2) and moderately
increased the insulation thickness for the exterior floor (EF1-T) rel-
viding the minimum operational energy in which a reduction of ative to the building’s initial design. Additionally, the solution
operational energy increases the embodied energy by up to 1911 implemented the passive house window (WN1-M), and used PIR
GJ (see Fig. 4, Min. OE). This indicates that there is a limit for the insulation for the concrete exterior wall (EW1-M) and the exterior
reduction of operational energy use by adopting energy efficiency floor (EF1-M), whereas all other material types (i.e. EW2-M and
measures beyond which further reductions can be unfavorable RF1-M) were kept the same as the initial design. These results fur-
from a life cycle energy perspective due to significant increases ther indicate that there is a limit to when additional reductions in
in the embodied energy use. operational energy through energy efficiency measures can be
Out of all the design parameters (i.e. energy efficiency mea- unfavorable from a life cycle energy perspective.
sures) considered in this study one, i.e. window type (WN1-M), There are also some solutions that simultaneously reduce both
converged to the same outcome (passive house window) for all the embodied and operational energy compared to the building’s
the Pareto solutions shown in Fig. 2. This indicates that although initial design, i.e. solutions located in the lower-left quadrant (neg-
the application of passive house windows increases the embodied ative energy differences) in Fig. 4. These solutions implemented
energy (by 123 GJ/m2 window, see Table 4) compared to a standard the passive house window (WN1-M), whereas all other material
window, the offsetting reduction in operational energy is more sig- types (i.e. EF1-M, EW1-M, EW2-M, and RF1-M) were kept the same
nificant which is favorable from a life cycle energy perspective. The as the initial design. Besides changing the window type, these
results thus suggest that the use of passive house windows is one solutions generally had thickness increases in the insulation of

8
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

Fig. 4. Results of the multi-objective optimization on the case building with respect to embodied and operational energy. The solutions in the figure are presented as the
difference in embodied and operational energy compared to the building’s initial design. The solutions that are selected to be subject to sensitivity analysis are also
highlighted and labeled. A color print for this figure is recommended.

the concrete exterior wall (EW1-T) and the roof (RF1-T), and thick- come for the discrete design parameters related to the material and
ness reductions in the insulation of the wood stud exterior wall component types (i.e. WN1-M, EF1-M, EW1-M, EW2-M, RF1-M).
(EW2-T) and exterior floor (EF1-T). Thereby, within these groups, the solutions differ only in terms
Based on the results of the multi-objective optimization shown of the values for the design parameters related to material quanti-
in Fig. 4 the Pareto solutions were divided into groups to provide ties (i.e. EF1-T, EW1-T, EW2-T, RF1-T). To obtain one solution from
the sensitivity analysis with a reduced set of Pareto solutions. A each group for further analysis, the solution with the median oper-
group was defined as a set of solutions that provided the same out- ational energy was chosen. Additionally, the compromise Pareto

Table 6
Design parameter values and results for embodied energy, operational energy, and life cycle energy for the selected Pareto solutions based on the grouping and selection process.

Solution WN1-M EF1- EF1-T EW1-M EW1-T EW2-M EW2-T RF1-M RF1-T D Embodied D Operational
M (m) (m) (m) (m) energy (GJ) energy (GJ)
1 Passive house EPS 0.06 Mineral 0.24 Mineral 0.09 Cellulose 0.60 –33 2148
window wool wool
2 Passive house PIR 0.06 Mineral 0.30 Mineral 0.25 Cellulose 0.60 137 3889
window wool wool
3 Passive house EPS 0.13 PIR 0.30 Mineral 0.25 Cellulose 0.60 524 4437
window wool
4 Passive house PIR 0.13 PIR 0.28 Mineral 0.25 Cellulose 0.60 521 4431
window wool
5 Passive house PIR 0.27 PIR 0.30 Mineral 0.25 Mineral 0.60 836 4627
window wool wool
6 Passive house EPS 0.08 PIR 0.30 PIR 0.23 Cellulose 0.60 1017 4653
window
7 Passive house PIR 0.15 PIR 0.30 PIR 0.25 Cellulose 0.60 1164 4934
window
8 Passive house PIR 0.24 PIR 0.30 PIR 0.25 Mineral 0.60 1407 5039
window wool
9 Passive house PIR 0.15 PIR 0.30 PIR 0.25 PIR 0.60 1771 5157
window
Min. Passive house PIR 0.16 PIR 0.30 Mineral 0.25 Cellulose 0.60 573 4520
LCE window wool
Min. EE Passive house PIR 0.06 Mineral 0.18 Mineral 0.01 Cellulose 0.43 129 106
window wool wool
Min. OE Passive house PIR 0.29 PIR 0.30 PIR 0.25 PIR 0.60 1911 5230
window

9
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

solutions, which here corresponds to the lowest life cycle energy detail in Table 6. The selected Pareto solutions were then subjected
(according to the scope presented in Fig. 1), and the Pareto solu- to a sensitivity analysis for supporting design decisions. The reason
tions with the minimum embodied and operational energy, respec- for performing sensitivity analysis on this range of Pareto solutions
tively, were selected. The selected Pareto solutions (12 solutions in (and not just focusing on those that commonly minimized embod-
total) are highlighted and labeled in Fig. 4 and described more in ied and operational energy) was to highlight how diverse optimal

Fig. 5. Results from the sensitivity analysis of the 12 selected Pareto solutions showing the absolute mean elementary effect m* with respect to the design parameters and
both embodied energy (EE) and operational energy (EE). A color print for this figure is recommended.

10
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

solutions (i.e. configurations of design parameters) can give differ- 4. Discussion and conclusion
ent results in terms of sensitivity.
The results from executing the sensitivity analysis on the 12 The use and integration of multi-objective optimization and
selected solutions presented in Table 6 are shown in Fig. 5, post-optimization sensitivity analysis were explored for finding
where the absolute mean elementary effect (m*) related to both and evaluating design solutions that consider the embodied/oper-
embodied and operational energy for the design parameters is ational energy trade-off in order to find solutions that favorably
presented. minimize the building’s life cycle energy use. For this purpose, an
In general, the results of the sensitivity analysis in Fig. 5 approach was developed and demonstrated in a case of a Swedish
show how the design parameters can exhibit different sensitivi- multifamily apartment building located in a subarctic climate.
ties to embodied and operational energy depending on the con- The sensitivity analysis approach proposed in this paper builds
figuration of the solution and its performance. The results of the on the previously recognized potential of sensitivity analysis to
sensitivity analysis indicate that for all selected solutions the provide valuable information about what design parameters to
thickness of the roof insulation (RF1-T, blue squares in Fig. 5) focus on and less influential design parameters that only have a
ranks the lowest or second lowest in terms of sensitivity to both minor impact on a building’s performance [33,34]. However, the
embodied and operational energy. This is especially clear in solu- proposed post-optimization sensitivity analysis is in contrast to
tions (i.e. solutions 1 to 7, Min. LCE and Min. EE) where the roof previous studies which mainly considered sensitivity analysis on
insulation material (RF1-M) converged to cellulose, which has a its own (e.g. [34,37]). Instead, the proposal presented in this paper
low embodied energy and high thermal conductivity compared is arranged such that sensitivity analysis is applied to solutions
to the other insulation materials (see Table 3). For solutions 9 that are generated using a multi-objective optimization approach.
and Min. OE where PIR insulation is used on the roof (RF1-M), The contribution of the proposed post-optimization sensitivity
the roof insulation thickness (RF1-T) resulted in modest sensitiv- analysis is that it can add valuable information that would comple-
ity to embodied energy due to higher embodied energy of PIR ment the results obtained using a multi-objective optimization
compared with other insulations. These results indicate that, in approach. Consequently, it can be used to support a decision-
general, due to the relatively low sensitivity of the roof insula- making process in the analysis and selection of solutions from a
tion thickness (RF1-T) to embodied and operational energy, its set of Pareto solutions to progress with, a task that is previously rec-
value can diverge more freely from the optimal value identified ognized as not always trivial [14]. Additionally, the type of post-
by the optimization process without significantly impacting optimization sensitivity analysis approach explored in this paper
either of the objectives. contrasts with, for example, robustness measure approaches which
The insulation thickness in the wood stud exterior wall (EW2-T, are used to calculate an aggregated metric to indicate how robust a
red circles in Fig. 5) indicated a high sensitivity to operational solution is to changes (e.g. [70]). These types of aggregated metrics
energy in all the selected solutions. This is partially an effect of can be useful as they provide a single metric that could potentially
the wood stud exterior wall having the largest enclosing area, be easier to interpret, however, the aim of the post-optimization
which is about twice the area of the concrete exterior wall sensitivity analysis is to expose individual design parameters for a
(EW1), and about three times the area of the exterior floor (EF1) deeper analysis. The outcome could thus provide a better under-
and roof (RF1). From an embodied energy and life cycle energy per- standing of the relationships and relative importance of specific
spective, the insulation thickness in the wood stud exterior wall design parameters and could provide valuable information for the
(EW2-T) has a higher sensitivity in solutions 6 to 9, and Min. OE, progression of the design going forward [33,34]. This process sup-
which commonly used PIR insulation. This indicates that, in gen- ports indications of what design parameters (1) have no or negligible
eral, when PIR insulation is selected as the optimal choice, it can influence on the objectives where deviations from their values and
strongly affect the building’s life cycle energy use depending on optimal choice could be made without significantly impacting the
its thickness and therefore needs to be selected carefully. As shown expected performance of the objectives, and (2) have a relatively
in Fig. 5, the results indicate that, in general, thicknesses of insula- substantial influence on the objectives and should thus not deviate
tion in both the exterior wall (i.e. EW1-T, EW2-T) result in higher significantly from their optimal values identified by the optimiza-
sensitivity in terms of embodied and operational energy compared tion process if the desired performance is to be maintained.
with the roof insulation thickness (RF1-T). This indicates that in Besides contributing to an approach for post-optimization sen-
contrast to the roof insulation thickness (as described above), the sitivity analysis, the proposed approach herein is based on the
insulation thicknesses in the exterior walls should be considered notion that the same processes for generating solutions (using
carefully and kept close to those of the identified value for each the parametric model) and evaluating them are used for both the
optimal solution (listed in Table 6) provided by the optimization multi-objective optimization and sensitivity analysis, respectively.
process. This has the practical benefit of not having to have separate pro-
The greatest variation in the sensitivity (from lowest to highest) cesses, which can be time-consuming to set up. Instead, if reuse
is found in the exterior floor insulation thickness (EF1-T, green tri- of already defined processes for multi-objective optimization
angles in Fig. 5), both in terms of embodied and operational energy. approaches can be performed, the addition of sensitivity analysis
For example, the exterior floor insulation thickness ranks the high- could become a more attractive proposal. This helps to better align
est for both embodied and operational energy in solution 2, indi- with Eisenhower et al. [31], who suggested that sensitivity analysis
cating that it is the most sensitive parameter. In contrast, it ranks should be integrated into any optimization process.
the lowest in terms of operational energy in solutions 5, 8, and The outcome of the case demonstration indicated the applica-
Min. OE. In general, the thickness of exterior floor insulation ranks bility of the approach proposed and developed in this study to suc-
higher when its nominal value is closer to the lower possible range cessfully identify the sensitivity of design parameters and obtained
of that parameter (see Table 2). The results thus indicate that for optimization results to aid supporting design decisions that favor-
the thicknesses lesser than 0.16 m, it is necessary to select the opti- ably minimize a building’s life cycle energy use. Thereby, multi-
mal exterior floor insulation close to the value that the optimiza- objective optimization was initially performed to identify a diverse
tion process converged to (see Table 6). In contrast, when aiming range of optimal Pareto solutions of various energy efficiency mea-
to use thicker insulations greater than 0.16 m in the exterior floor sures that could result in life cycle energy savings of up to 4520 GJ
its value can, due to the lower sensitivity, be selected with a higher relative to the case building’s initial design. Later, 12 Pareto solu-
degree of freedom.
11
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

tions were selected and subjected to a sensitivity analysis for sup- cycle energy assessment in this research did not cover the energy
porting design decisions. The results of this sensitivity analysis used in construction and end-of-life phases, which should be
indicated that due to the relatively low sensitivity of the roof insu- expanded upon in future studies.
lation thickness to embodied and operational energy, its value
could be modified and selected more freely without significantly Declaration of Competing Interest
impacting the building’s life cycle energy use. Thicknesses of insu-
lation in exterior walls showed in general higher sensitivity than The authors declare that they have no known competing finan-
the roof insulation thickness and therefore in contrast to the roof, cial interests or personal relationships that could have appeared
their value needs to be selected carefully and kept close to those to influence the work reported in this paper.
identified by the optimization. The thickness of exterior floor insu-
lation exposed the greatest variation in the sensitivity which Acknowledgements
ranked higher when its nominal value is closer to the lower possi-
ble range of that parameter. The results thus indicated that for the This work was supported by the Swedish Energy Agency
exterior floor thicknesses lesser than 0.16 m, it is necessary to through the E2B2 project (grant number 2019-020708); and the
select the optimal insulation thickness close to the value that the Swedish research council Formas (grant number 2018-01463).
optimization process converged to, while, when aiming to use The authors would also like to thank the NCC construction com-
thicker insulations greater than 0.16 m its optimal value can be pany for providing data and feedback relating to the building case
selected more freely due to the lower sensitivity in terms of studied.
embodied and operational energy.
From a practical perspective, applications of multi-objective References
optimization are currently often limited to consider two or three
objectives and a relatively limited number of constraints [14], as [1] United Nations Environment Programme. 2020 Global Status Report for
Buildings and Construction: Towards a Zero-emission, Efficient and Resilient
is the case herein. Therefore, having these sensitivity indices could Buildings and Construction Sector. https://globalabc.org/sites/default/files/
also be useful when an optimal solution is considered in light of inline-files/2020%20Buildings%20GSR_FULL%20REPORT.pdf. Accessed 04/01,
other criteria (e.g. constructability, logistics, procurement, etc.) 2021.
[2] European Parliament and European Council. Directive (EU) 2018/844 of the
that were not explicitly included through design parameters, European Parliament and of the Council of 30 May 2018 amending Directive
objectives, or constraints in the optimization process. Including 2010/31/EU on the energy performance of buildings and Directive 2012/27/EU
such considerations when evaluating an optimal solution might on energy efficiency. Off. J. Eur. Union (2018).
[3] McGraw-Hill Construction. World green building trends: Business benefits
result in changes needed to the configuration of that optimal solu-
driving new and retrofit market opportunities in over 60 countries. Bedford
tion. The potential effect of such changes on the objectives and Massachusetts: Smart Market Report, 2013.
expected performance of the solution could be qualitatively evalu- [4] M.M. Khasreen, P.F.G. Banfill, G.F. Menzies, Life-cycle assessment and the
ated using their sensitivity provided by the post-optimization sen- environmental impact of buildings: a review, Sustainability 1 (2009) 674–701,
https://doi.org/10.3390/su1030674.
sitivity analysis. [5] C.J. Kibert, Sustainable Construction: Green Building Design and Delivery, John
Wiley & Sons, 2016.
[6] A. Schlueter, F. Thesseling, Building information model based energy/exergy
performance assessment in early design stages, Autom. Constr. 18 (2) (2009)
5. Limitations and future research 153–163, https://doi.org/10.1016/j.autcon.2008.07.003.
[7] M. Rekola, T. Mäkeläinen, T. Häkkinen, The role of design management in the
The scope of this study was chosen to be focused on the trade- sustainable building process, Architectural Eng. Design Manage. 8 (2) (2012)
78–89, https://doi.org/10.1080/17452007.2012.659503.
off between embodied/operational energy as it provides a demon- [8] P. Chastas, T. Theodosiou, D. Bikas, Embodied energy in residential buildings-
stration for the purpose of exploring how multi-objective opti- towards the nearly zero energy building: a literature review, Build. Environ.
mization and post-optimization sensitivity analysis could be 105 (2016) 267–282, https://doi.org/10.1016/j.buildenv.2016.05.040.
[9] M.K. Dixit, Life cycle embodied energy analysis of residential buildings: a
used. However, future studies need to extend this scope to consider review of literature to investigate embodied energy parameters, Renew.
additional, or other, trade-off scenarios that are relevant to build- Sustain. Energy Rev. 79 (2017) 390–413, https://doi.org/10.1016/j.
ing design. This could include, for example, life cycle carbon rser.2017.05.051.
[10] J. Kneifel, E. O’Rear, D. Webb, C. O’Fallon, An exploration of the relationship
impact, life cycle cost, and different design contexts (building
between improvements in energy efficiency and life-cycle energy and carbon
types, geographic locations, etc.). emissions using the BIRDS low-energy residential database, Energy Build. 160
The proposed approach in this study also only considered a sub- (2018) 19–33, https://doi.org/10.1016/j.enbuild.2017.11.030.
set of a design process and it is suggested that it should be inte- [11] D. Tuhus-Dubrow, M. Krarti, Genetic-algorithm based approach to optimize
building envelope design for residential buildings, Build. Environ. 45 (7)
grated with other approaches, defining a more holistic (2010) 1574–1581, https://doi.org/10.1016/j.buildenv.2010.01.005.
proposition. This can include, for example, factor fixing to screen [12] I. Caetano, L. Santos, A. Leitão, Computational design in architecture: defining
out insignificant design parameters prior to performing optimiza- parametric, generative, and algorithmic design, Front. Architectural Res. 9 (2)
(2020) 287–300, https://doi.org/10.1016/j.foar.2019.12.008.
tion to reduce the computational time [14,37] and techniques for [13] R. Evins, A review of computational optimisation methods applied to
supporting multi-criteria decision making [49]. Additionally, the sustainable building design, Renew. Sustain. Energy Rev. 22 (2013) 230–245,
Method of Morris was chosen as the sensitivity analysis method, https://doi.org/10.1016/j.rser.2013.02.004.
[14] A.T. Nguyen, S. Reiter, P. Rigo, A review on simulation-based optimization
however, other methods could offer additional insights and higher methods applied to building performance analysis, Appl. Energy 113 (2014)
accuracy (such as variance-based methods [39]), and further stud- 1043–1058, https://doi.org/10.1016/j.apenergy.2013.08.061.
ies should investigate the selection of appropriate methods. [15] F. Kheiri, A review on optimization methods applied in energy-efficient
building geometry and envelope design, Renew. Sustain. Energy Rev. 92 (2018)
In regard to the evaluation of a building’s life cycle energy per- 897–920, https://doi.org/10.1016/j.rser.2018.04.080.
formed in this study, there are some other parameters that have [16] V. Machairas, A. Tsangrassoulis, K. Axarli, Algorithms for optimization of
potentially considerable effects on the results, such as the scope building design: a review, Renew. Sustain. Energy Rev. 31 (2014) 101–112,
https://doi.org/10.1016/j.rser.2013.11.036.
and system boundary of the life cycle energy assessment, the
[17] Z. Jalali, E. Noorzai, S. Heidari, Design and optimization of form and façade of
building’s lifespan (assumed to be 50 years herein), and the an office building using the genetic algorithm, Sci. Technol. Built Environ. 26
embodied energy data used for different materials and compo- (2) (2020) 128–140, https://doi.org/10.1080/23744731.2019.1624095.
nents. The impact of these factors was not considered herein, how- [18] D. D’Agostino, P. D’Agostino, F. Minelli, F. Minichiello, Proposal of a new
automated workflow for the computational performance-driven design
ever, they should be taken into consideration in future research optimization of building energy need and construction cost, Energy Build.
works. Additionally, the scope of the embodied energy and life 239 (2021) 110857, https://doi.org/10.1016/j.enbuild.2021.110857.

12
J. Mukkavaara and F. Shadram Energy & Buildings 253 (2021) 111529

[19] F. Rosso, V. Ciancio, J. Dell’Olmo, F. Salata, Multi-objective optimization of [44] N. Emami, Untangling parameters: a formalized framework for identifying
building retrofit in the Mediterranean climate by means of genetic algorithm overlapping design parameters between two disciplines for creating an
application, Energy Build. 216 (2020) 109945, https://doi.org/10.1016/j. interdisciplinary parametric model, Adv. Eng. Inf. 42 (2019) 100943, https://
enbuild.2020.109945. doi.org/10.1016/j.aei.2019.100943.
[20] F. Ascione, N. Bianco, G.M. Mauro, G.P. Vanoli, A new comprehensive [45] R. Yu, J. Gero. An Empirical Foundation for Design Patterns in Parametric
framework for the multi-objective optimization of building energy design: Design. In Proceedings of the 20th International Conference of the Association
Harlequin, Appl. Energy 241 (2019) 331–361, https://doi.org/10.1016/j. for Computer-Aided Architectural Design Research in Asia (CAADRIA), 2015.
apenergy.2019.03.028. [46] A. Borrmann, V. Berkhahn. Principles of Geometric Modeling. In: A. Borrmann,
[21] M. Ghaderian, F. Veysi, Multi-objective optimization of energy efficiency and M. König, C. Koch, J. Beetz. (eds). Building Information Modeling. Springer,
thermal comfort in an existing office building using NSGA-II with fitness 2018, doi: 10.1007/978-3-319-92862-3_2.
approximation: a case study, J. Build. Eng. 41 (2021) 102440, https://doi.org/ [47] T. Cerovsek, A review and outlook for a ‘Building Information Model’ (BIM): A
10.1016/j.jobe.2021.102440. multi-standpoint framework for technological development, Adv. Eng. Inf. 25
[22] M. Pazouki, K. Rezaie, A. Bozorgi-Amiri, A fuzzy robust multi-objective (2) (2011) 224–244, https://doi.org/10.1016/j.aei.2010.06.003.
optimization model for building energy retrofit considering utility function: [48] R. Sacks, C.M. Eastman, G. Lee, Parametric 3D modeling in building
A university building case study, Energy Build. 241 (2021) 110933, https://doi. construction with examples from precast concrete, Autom. Constr. 13 (3)
org/10.1016/j.enbuild.2021.110933. (2004) 291–312, https://doi.org/10.1016/S0926-5805(03)00043-8.
[23] Y. Jung, Y. Heo, H. Lee, Multi-objective optimization of the multi-story [49] M. Sandberg, J. Mukkavaara, F. Shadram, T. Olofsson, Multidisciplinary
residential building with passive design strategy in South Korea, Build. optimization of life-cycle energy and cost using a BIM-based master model,
Environ. 203 (2021) 108061, https://doi.org/10.1016/j.buildenv.2021.108061. Sustainability 11 (1) (2019) 286, https://doi.org/10.3390/su11010286.
[24] A. Ciardiello, F. Rosso, J. Dell’Olmo, V. Ciancio, M. Ferrero, F. Salata, Multi- [50] K. Deb, Multi-objective genetic algorithms: Problem difficulties and
objective approach to the optimization of shape and envelope in building construction of test problems, Evol. Comput. 7 (3) (1999) 205–230, https://
energy design, Appl. Energy 280 (2020) 115984, https://doi.org/10.1016/j. doi.org/10.1162/evco.1999.7.3.205.
apenergy.2020.115984. [51] K. Deb. Multi-objective optimisation using evolutionary algorithms: An
[25] B. Chegari, M. Tabaa, E. Simeu, F. Moutaouakkil, H. Medromi, Multi-objective introduction. Springer, 2011, doi: 10.1007/978-0-85729-652-8_1.
optimization of building energy performance and indoor thermal comfort by [52] G. Hammond, C. Jones. Inventory of carbon and energy (ICE), Version 2.0.
combining artificial neural networks and metaheuristic algorithms, Energy Sustainable energy research team, University of Bath, https://
Build. 239 (2021) 110839, https://doi.org/10.1016/j.enbuild.2021.110839. www.circularecology.com/embodied-energy-and-carbon-footprint-database.
[26] U. Acar, O. Kaska, N. Tokgoz, Multi-objective optimization of building envelope html. Accessed 04/01, 2021.
components at the preliminary design stage for residential buildings in [53] Norwegian EPD foundation (EPD Norway). http://epd-norge.no/epder/.
Turkey, J. Build. Eng. 42 (2021) 102499, https://doi.org/10.1016/j. Accessed 04/01, 2021.
jobe.2021.102499. [54] The International EPDÒ System-Environmental Product Declarations. http://
[27] F. Shadram, J. Mukkavaara, An integrated BIM-based framework for the environdec.com/. Accessed 04/01, 2021.
optimization of the trade-off between embodied and operational energy, Energy [55] Swedish building code, Swedish national board of housing, building and
Build. 158 (2018) 1189–1205, https://doi.org/10.1016/j.enbuild.2017.11.017. planning (Boverket). Regulations and general advice on accessibility, housing
[28] F. Shadram, J. Mukkavaara, Exploring the effects of several energy efficiency design, room height, operating space, fire protection, hygiene, health and
measures on the embodied/operational energy trade-off: a case study of environment, noise protection, safety in use and energy conservation. https://
swedish residential buildings, Energy Build. 183 (2019) 283–296, https://doi. www.boverket.se/contentassets/a9a584aa0e564c8998d079d752f6b76d/
org/10.1016/j.enbuild.2018.11.026. konsoliderad_bbr_2011-6.pdf. Accessed 04/01, 2021.
[29] S. Abbasi, E. Noorzai, The BIM-Based multi-optimization approach in order to [56] J. Blank, K. Deb, pymoo: multi-objective optimization in python, IEEE Access 8
determine the trade-off between embodied and operation energy focused on (2020) 89497–89509, https://doi.org/10.1109/ACCESS.2020.2990567.
renewable energy use, J. Cleaner Prod. 281 (2021) 125359, https://doi.org/ [57] E. Triantaphyllou, E. Multi-criteria decision making methods. Springer, 2000,
10.1016/j.jclepro.2020.125359. doi: 10.1007/978-1-4757-3157-6_2.
[30] K. Deb, A. Pratap, S. Agarwal, T.A.M.T. Meyarivan, A fast and elitist [58] H.A. Taboada, F. Baheranwala, D.W. Coit, N. Wattanapongsakorn, Practical
multiobjective genetic algorithm: NSGA-II, IEEE Trans. Evol. Comput. 6 (2) solutions for multi-objective optimization: an application to system reliability
(2002) 182–197, https://doi.org/10.1109/4235.996017. design problems, Reliab. Eng. Syst. Saf. 92 (3) (2007) 314–322, https://doi.org/
[31] B. Eisenhower, Z. O’Neill, S. Narayanan, V.A. Fonoberov, I. Mezić, A 10.1016/j.ress.2006.04.014.
methodology for meta-model based optimization in building energy models, [59] W. Tian, A review of sensitivity analysis methods in building energy analysis,
Energy Build. 47 (2012) 292–301, https://doi.org/10.1016/j. Renewable Sustainable Energy Rev 20 (2013) 411–419, https://doi.org/
enbuild.2011.12.001. 10.1016/j.rser.2012.12.014.
[32] A. Saltelli, S. Tarantola, F. Campolongo, M. Ratto, Sensitivity Analysis in [60] M.D. Morris, Factorial sampling plans for preliminary computational
Practice, John Willey & Sons (2004). experiments, Technometrics 33 (2) (1991) 161–174.
[33] A.T. Nguyen, S. Reiter, A performance comparison of sensitivity analysis [61] F. Campolongo, J. Cariboni, A. Saltelli, A. An effective screening design for
methods for building energy models, Build. Simul. 8 (6) (2015) 651–664, sensitivity analysis of large models. Environ. Modelling Software 22(10)
https://doi.org/10.1007/s12273-015-0245-4. (2007) 1509-1518, doi: 10.1016/j.envsoft.2006.10.004.
[34] P. Heiselberg, H. Brohus, A. Hesselholt, H. Rasmussen, E. Seinre, S. Thomas, [62] J. Herman, W. Usher, SALib: An open-source Python library for sensitivity
Application of sensitivity analysis in design of sustainable buildings, analysis, J. Open Source Software 2 (9) (2017), https://doi.org/10.21105/
Renewable Energy 34 (9) (2009) 2030–2036, https://doi.org/10.1016/j. joss.00097.
renene.2009.02.016. [63] Y. Yildiz, K. Korkmaz, T. Göksal Özbalta, Z. Durmus Arsan, An approach for
[35] C.J. Hopfe, J.L. Hensen, Uncertainty analysis in building performance developing sensitive design parameter guidelines to reduce the energy
simulation for design support, Energy Build. 43 (10) (2011) 2798–2805, requirements of low-rise apartment buildings, Appl. Energy 93 (2012) 337–
https://doi.org/10.1016/j.enbuild.2011.06.034. 347, https://doi.org/10.1016/j.apenergy.2011.12.048.
[36] M.H. Kristensen, S. Petersen, Choosing the appropriate sensitivity analysis [64] F. Ascione, N. Bianco, T. Iovane, G.M. Mauro, D.F. Napolitano, A. Ruggiano, L.
method for building energy model-based investigations, Energy Build. 130 Viscido, A real industry building: Modeling, calibration and Pareto
(2016) 166–176, https://doi.org/10.1016/j.enbuild.2016.08.038. optimization of energy retrofit, J. Build. Eng. 29 (2020), https://doi.org/
[37] T. Østergård, R.L. Jensen, S.E. Maagaard, Early Building Design: Informed 10.1016/j.jobe.2020.101186 101186.
decision-making by exploring multidimensional design space using sensitivity [65] F. Ascione, N. Bianco, C. De Stasio, G.M. Mauro, G.P. Vanoli, Multi-stage and
analysis, Energy Build. 142 (2017) 8–22, https://doi.org/10.1016/j. multi-objective optimization for energy retrofitting a developed hospital
enbuild.2017.02.059. reference building: A new approach to assess cost-optimality, Appl. Energy
[38] Z. Pang, Z. O’Neill, Y. Li, F. Niu, The role of sensitivity analysis in the building 174 (2016) 37–68, https://doi.org/10.1016/j.apenergy.2016.04.078.
performance analysis: a critical review, Energy Build. 209 (2020) 109659, [66] CRAN – Package sensitivity. https://cran.r-project.org/web/packages/
https://doi.org/10.1016/j.enbuild.2019.109659. sensitivity/index.html. Accessed 09/13, 2021.
[39] D.G. Sanchez, B. Lacarrière, M. Musy, B. Bourges, Application of sensitivity [67] A. Berge, P. Johansson, Literature Review of High Performance Thermal
analysis in building energy simulations: combining first-and second-order Insulation, Chalmers University of Technology, 2012.
elementary effects methods, Energy Build. 68 (2014) 741–750, https://doi.org/ [68] C. Hill, A. Norton, J. Dibdiakova, A comparison of the environmental impacts of
10.1016/j.enbuild.2012.08.048. different categories of insulation materials, Energy Build. 162 (2018) 12–20,
[40] C. Spitz, L. Mora, E. Wurtz, A. Jay, Practical application of uncertainty analysis https://doi.org/10.1016/j.enbuild.2017.12.009.
and sensitivity analysis on an experimental house, Energy Build. 55 (2012) [69] Sveby, standards for energy in buildings. Input data for calculation of
459–470, https://doi.org/10.1016/j.enbuild.2012.08.013. building’s energy use. http://www.sveby.org/wp-content/uploads/2012/10/
[41] A. Saltelli, M. Ratto, T. Andres, F. Campolongo, J. Cariboni, D. Gatelli, M. Sveby_Brukarindata_bostader_version_1.0.pdf. Accessed 04/01, 2021.
Saisana, S. Tarantola (Eds.), Global Sensitivity Analysis. The Primer, John Wiley [70] P. Hoes, M. Trcka, J.L.M. Hensen, B.H. Bonnema, Optimizing building designs
& Sons, Ltd, Chichester, UK, 2007. using a robustness indicator with respect to user behavior, in: Building
[42] J. Zajas, P. Heiselberg, Parametric study and multi objective optimization of simulation proceedings of the 12th conference of the international building
window frame geometry, Build. Simul. 7 (6) (2014) 579–593. performance simulation association, 2011, pp. 1710–1717.
[43] J. Mukkavaara, M. Sandberg. Architectural design exploration using generative
design: framework development and case study of a residential block.
Buildings 10(11) (2020) 201, doi: 10.3390/buildings10110201.

13

You might also like