See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/325425422

Clustering Forms for Enhancing Architectural Design Optimization

Conference Paper · May 2018

DOI: 10.52842/conf.caadria.2018.2.431

CITATIONS READS

5 1,798

2 authors:

Shermeen Yousif Wei Yan

Florida Atlantic University Texas A&M University
13 PUBLICATIONS 55 CITATIONS 59 PUBLICATIONS 1,407 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Shermeen Yousif on 29 May 2018.

The user has requested enhancement of the downloaded file.

CLUSTERING FORMS FOR ENHANCING ARCHITECTURAL
DESIGN OPTIMIZATION

SHERMEEN YOUSIF1 and WEI YAN2

1,2
Texas A&M University
1,2
{shermeen|wyan}@tamu.edu

Abstract. This work introduces a new system in architectural design

optimization that integrates form diversity and clustering methods
into the process. The first method we propose is an algorithm for
rating design solutions according to their geometric correspondences,
maximizing differences and enforcing diversity. In addition, we
implement the K-means algorithm to cluster the resulting design
forms into groups of similar forms, to substitute each group with one
representative solution. The work aims to facilitate decision making
and form evaluation for designers, leading to an interactive optimization
process, and contributing to improving existing optimization models
in architectural design research and practice. We modeled a dynamic
system through prototyping, experimenting and test-case application.
As a prototype development, the protocol was done through phases
of: (1) parametric modeling, (2) conducting energy simulation and
daylight analysis and running a generative system, and (3) developing
an algorithm for form diversity and another for implementing K-means
clustering. The results are illustrated and discussed in detail in the paper.

Keywords. Architectural Design Optimization; Form Diversity;

K-Means Clustering.

1. Introduction
Architectural design can be treated as an optimization process in which finding
optimal solutions that satisfy predetermined objectives is targeted (Yan et al.
2015). A continuous thread of research has been conducted on architectural
design optimization and generative systems to evolve performance-driven building
design solutions. Despite the potential in advancing the design process of
efficient buildings, optimization models still suffer multiple issues (Attia et al.
2013). One issue in available optimization platforms is that appraisal of form
qualities is not explicitly incorporated in the process. Design optimization has
to integrate building form evaluation in addition to other performance-based
criteria to guarantee that a full definition of a building is developed (Yi et
al. 2015). Another issue is the too many, and sometimes redundant design
solutions often generated from the search mechanism (e.g. hundreds), and
they can be similar in form due to the possibility of slight changes in model’s
parameters. The overwhelming number of solutions causes designers’ fatigue,

T. Fukuda, W. Huang, P. Janssen, K. Crolla, S. Alhadidi (eds.), Learning, Adapting and Prototyping,
Proceedings of the 23rd International Conference of the Association for Computer-Aided Architectural
Design Research in Asia (CAADRIA) 2018, Volume 2, 431-440. © 2018 and published by the Association
for Computer-Aided Architectural Design Research in Asia (CAADRIA) in Hong Kong.
432 S. YOUSIF AND W. YAN

inhibits effective decision making, and leads to inefficient interaction between the
designers and the optimization system. This work introduces an innovative system
in architectural design optimization that integrates form diversity, facilitating
architects’ evaluation of design solution forms. The first objective is to develop an
algorithm for form diversity to rate design solutions according to their geometric
differences, and to exclude similar forms. Another solution to the excessive
number of solutions is to cluster the design solutions resulting from optimization
into groups. This enables designers to perceive the grouped forms, and to have a
control on manipulating the number of clusters (to decrease the number) of final
design solutions. The primary contribution of this work is a simple, yet robust
algorithm for finding correspondences between geometric forms, maximizing
difference sand enforcing diversity. The aim is to apply this algorithm to other
forms, more complex test-cases, and test the potential generalization of the
method. Ultimately, the contribution is to computational design methods through
accommodating designers’ aesthetics and form appraisal into architectural design
optimization.

2. Background
Recently, progress towards developing optimization methods has been
investigated and integrated into architectural design research (Aish et al. 2012).
Several studies have been published in the field that proposed or experimented
with the use of optimization to create building form alternatives(Caldas and
Norford 2003, Gerber et al. 2012, Welle et al. 2012, Bradner and Davis
2013, Konis et al. 2016). For Building (2D) shape optimization, Wang et
al. (2006) have employed a Multi-Objective Genetic Algorithm (MOGA) to
optimize building layout and orientation. A commonality shared by these studies
is the focus on producing variations of building forms, informed mainly by
energy footprint or/and other quantifiable criteria, without considering issues
of form diversity or the pursuit of eliminating similar forms. The system we
suggest here includes tasks in the order of (1) parametric modeling, (2) energy
simulation and daylight analysis, to be run during optimization, (3) in addition to
post-optimization articulation of solutions utilizing form diversity and K-means
clustering algorithms. The following is an explanation of technical areas and
methods that are utilized in this work.

2.1. PARAMETRIC MODELING

Parametric Modeling is a method to define models with constraints and variable
parameters utilizing rules to automatically modify design options based on
changing parameters (Aish and Woodbury 2005). One of the most intelligent
uses of parametric modeling is related to exploring new forms, and allowing for
a wide range of building efficiencies, such as structure, energy, construction, and
circulation (Caplan 2011).
 CLUSTERING FORMS FOR ENHANCING ARCHITECTURAL DESIGN 433
OPTIMIZATION

2.2. ENERGY SIMULATION AND DAYLIGHT STUDY

In a performance-based design approach, energy simulation typically needs to be
coupled with daylight studies for an accurate whole building simulation (Roudsari
et al. 2013). Simulating every design solution, a maximized preferred daylight
measure and minimum energy use are targeted in this study as the quantifiable
performance objectives that need to be satisfied.

2.3. MULTI-OBJECTIVE EVOLUTIONARY ALGORITHM

To solve Multi-Objective Optimization (MOO) problems, two methods are
typically followed. The first is the weighted sum method which is to utilize
a composite function of the sum of all objectives, each given a weight. The
second method is the Pareto Front, which is to find the set of non-inferior optimal
solutions (Radford and Gero 1987, Konak et al. 2006). A Pareto optimal solution
is retrieved through an optimal trade-off between two or more objectives, where
an objective cannot be improved without the other being degraded (Vierlinger
2013). A Multi-Objective Evolutionary Algorithm (MOEA) approach is a method
to solve optimization of multiple objectives utilizing the principles of evolution,
through search space exploration in a generative system to find the fittest solutions
(Vierlinger 2013). This work utilizes a MEOA-based tool for Pareto optimization.

2.4. FORM DIVERSITY

Often, genetic diversity is used to penalize closely-distanced solutions in the
solution space to escape the local optima. Genetic diversity is a distance-based
measure that can be often calculated through measuring the Euclidean distances
among all individuals in a population in the solution space; the diversity metric of
one individual is the aggregation of its distances from all of the others (Toffolo and
Benini 2003). In the work of (Brown et al. 2015), a structural design optimization
model, the search for diversity in solutions and seeking interaction with the
performance feedback has been made utilizing genetic diversity. Unlike those
works that focus on the genetic diversity of solutions’ parameters in the genotypical
space during optimization (which can be dimensions, materials, room connectivity,
etc.), our work is focused on diversity of forms by comparing the geometric
correspondences among the candidates post to optimization, considering only
the form parameters. For instance, when a rotation of a design solution is a
parameter, multiple rotated solutions mean multiple different solutions (different
in parameters) in terms of genetic diversity, while they all mean one form in
terms of our form diversity. In a prior work, a form diversity algorithm has been
developed through creating a program to rate resulting forms according to their
geometric differences, as the algorithm discards similar forms, and retains highly
diverse candidates (Yousif et al. 2017).

2.5. K-MEANS CLUSTERING

The concept of cluster analysis becomes important for understanding meaningful
groups that share mutual characteristics; collectively, cluster analysis is important
for variety of areas, such as pattern recognition (Tan 2006, Tan et al. 2013).
434 S. YOUSIF AND W. YAN

K-means clustering is an algorithm to divide a matrix of observations or data(like

points) in N dimensions into K clusters in which the within-cluster sum of squared
errors is minimized (Hartigan and Wong 1979). The main steps of K-means
algorithm can be explained as follows: (1) selecting the initial division with K
clusters, and repeat steps 2 and 3 until cluster membership stabilizes,(2) creating
a new partition by assigning each point to its closest cluster center, and (3)
computing new cluster centers (Jain and Dubes 1988, Jain 2010). The rationale
behind using the K-means clustering method in our work is because it is a
promising approach to group the design solutions into sets of similar forms, then
to find the representative solution candidate of each cluster to be presented to
designers, eliminating similarity, and reducing the number of final solutions. In
concept, in our research, for a set of design solutions in the solution space of a
generative system, the search is for clusters (the set of solutions that belong to each
cluster), analyzed according to their parameters of geometric forms. In addition,
it would be beneficial to find the centroid of each cluster, then the representative
candidate as the closest solution to that centroid (Kanungo et al. 2002). Although
K-means utilizes the Euclidean distance measure, we apply it only on geometric
parameters after optimization, as explained in the following section.

3. Methods
To illustrate the system we suggest, we created a dynamic model of the system
workflow, through prototyping, experimenting, and test-case application. In
prototyping, a series of procedures and steps were conducted and tested to develop
a functional prototype for optimization, and for developing and implementing the
form diversity and K-means clustering methods.

3.1. PROTOTYPE DEVELOPMENT

We followed the protocol illustrated below in Figure 1 for prototype
development.In phase one, we set up the initial model using parametric modeling,
in Grasshopper/Rhino as the main platform. Next, we prepare for energy
simulation and daylight analysis utilizing Honeybee and Ladybug plugins, and
conduct optimization of those objectives using the Multi-Objective Evolutionary
Algorithm tool (Octopus) as the search mechanism. In the third phase, post to
optimization, we created and implemented a form comparison algorithm using
Python and programmed an algorithmic definition to apply K-means clustering
analysis on resulting forms, using the Owl plugin nodes. The prototype has been
developed considering early design phase, although it can be applied to other
phases. Multiple tools were used for the prototype, as indicated in Figure 1.

3.2. TEST-CASE EXPERIMENT

To illustrate the prototype, we have utilized a test-case in which we conducted
thefollowing processes:
 CLUSTERING FORMS FOR ENHANCING ARCHITECTURAL DESIGN 435
OPTIMIZATION

3.2.1. Creating a Grid-Based Pattern

In a previous work, a simple grid-based pattern was created that relies on variable
allocation of units (boxes) to a cellular grid, allowing for 4.5913309e+20 options
to be generated, creating patterns such as in Figure 2. The rationale for using
such a heavily restricted grid-based layout is due to its simplicity yet capability
to generate a wide range of design options (Michalek et al. 2002), that are
required for testing the form diversity algorithm (Yousif et al. 2017). Another
reason for utilizing this layout is its capability to produce typological design
options. Typological designs refer to designs that represent typical typologies of
building configurations such as L-shapes and U-shapes, considering the fact that
designers do not utilize pure typologies, rather mixtures of typologies (Wurzer et
al. 2017). Our layout was set up using a grid of (11x11) cell, in which 10 units
(boxes) are allowed to occupy the grid. The first unit was allocated in the center
of the grid(x=6, y=6), and 9 units were appended consequently, allowing their
center points to changes as parameters. By coding customized definitions, three
algorithmic constraints were used in this layout: (1) adjacency, each unit must be
adjacent to at least one side of the previous unit/s, (2) non-overlap, a unit cannot be
located on an occupied cell, and (3) boundary-respect, the grid boundary has to be
satisfied, and units cannot be appended outside the grid. Samples of the resulted
forms are illustrated in Figure 2. The grid-based setup was meant to be a simplified
design form to test and experiment with the form diversity and K-means cluster
analysis, acknowledging that building designs are often more complex.

3.2.2. Setup Performance Evaluation and Optimization Run

To conduct performance simulation while the MOEA runs, a preliminary task is
needed, in which the setup of simulation requirements is made. To prepare for
energy simulation, the geometry of the initial model was converted into zones,
and a program of an office space was attributed to the zones. Next, glazing ratios
were selected to each façade, with 60% Window-To-Wall (WWR) ratio for the
north and south facades and 0% WWR for the east and west façades. Weather
file of College Station, TX (TMY3 format), simulation time-step, and schedules
were added to the energy simulation definition. Cooling loads were selected to be
the main loads to minimize for the energy fitness function because it is typically
challenging to minimize the cooling loads while maximizing daylight illuminance
for the hot humid climate selected here as the two criteria conflicts. For the
daylight study, materials were added to prescribe the building components: walls,
floors, and ceilings. The spatial Daylight Autonomy (sDA), which is a measure of
illuminance levels that needs to be achieved for LEED v4, was selected as a metric;
it has to be between(300-3000) lux for 9 a.m. and 3 p.m., both on a clear-sky day
at the equinox (USGB 2014). A grid of daylight sensors was created according to
the floor plate of the initial model; the sensors were distributed 5 meters apart, to
save computation time. A standard sky generator in addition to an exterior surface,
adjacent to the building model were connected to the grid-based simulation. The
maximum ratio of the area that achieves 300-3000 lux over the total area was
pursued as the fitness value.
436 S. YOUSIF AND W. YAN

To run the simulation-based MOEA algorithm, the mass model, the 9

parameters, and the fitness values (minimum cooling loads, and maximum
sDA ratio) were connected to the optimization tool (Octopus). For every
generation,energy simulation and daylight simulation are conducted for all design
candidates(evaluation pool size was 191). In some solutions, the form pattern
took north-south orientation to maximize daylight illuminance (e.g. solution 2)
which the solution has successfully achieved, while it is considered one of the
solutions of high cooling loads. In other cases, forms are east-west oriented
with minimum windows to minimize coaling loads (e.g. solution 1) with low
illuminance values, as shown in Figure 3. The Octopus solution space, illustrated
in Figure 4, shows the results of optimization at generation 20. The Pareto front is
indicated as a curve with 8 solutions, represented as meshes, and the Elite solutions
(the best solutions to potentially survive in the next generations) are also shown as
meshes, while solutions of previous generations are represented as dots (history).
All of the Pareto front, and 12 of the Elite solutions were marked (in bubbles)
and reinstated into Grasshopper for K-means clustering. The optimization results
of this experiment confirm an expected behavior to satisfy the two conflicting
objectives, and prove the algorithmic run successful, with the results of solution 1
and 2 as examples.

3.2.3. Develop and Implement Form Diversity Algorithm and K-means Clustering

• 3.2.3.1. Form Diversity Algorithm

In a prior work, post to optimization, a hundred of the Pareto front and the Elite
solutions were retrieved from generation 10 of a different optimization experiment
and were subjected to form comparison. From 100 solutions, the algorithm filtered
20 of high difference values, thus of high form diversity (Yousif et al. 2017). To
compute form diversity as a measure, the form comparison algorithm was created
to perform four procedures as follows. In (1) pairwise-comparison & translation
(move), every design solution was paired with all of the other 99 solutions, and
for each pairwise comparison, 10 translation operations were made to find the
maximum overlap and the minimum difference value. Step (2) is to record the
form difference values; the minimum difference value was considered as the actual
difference value for each pair compared, and thus was recorded. Next, in (3)
cross-reference matrix, the difference values between a solution and the other
99 were stored for each solution in a cross-reference matrix. Finally, in step (4)
filtering solutions of high difference values occurred, where solutions with higher
difference values were retained and solutions with lower difference values were
discarded (Yousif et al. 2017).
• 3.2.3.2. K-means Clustering Algorithm
At generation 20, the Pareto front (8 solutions) and 12 selected Elite solutions,
retrieved from optimization, were subjected to multiple procedures to prepare
the required parameters to the K-means algorithmic definition. The K-means
algorithm from the Owl tool requires tensors as the primary parameter, in addition
to the number of clusters, number of K-means iterations (was set to 100), and
 CLUSTERING FORMS FOR ENHANCING ARCHITECTURAL DESIGN 437
OPTIMIZATION

seeds (was set to 2399). Tensors refer to mathematical objects, used to describe
physical properties similar to scalars and vectors; a scalar is a zero rank tensor,and
a vector is a first rank tensor (Kline 1990). In the Owl tool, according to the
author, the K-means algorithm was created to accommodate multiple dimensions
or coordinates, N (numbers) of dimensions, and in our test-case we have 10
dimensions representing the 10 center points of each design solution. The next
required task was to convert the ten center points of the ten units of each design
solution into tensors in 10-dimensional K-means clustering. As a result, the
K-means definition provided the clusters, and indices of the list items that belong
to each cluster. It is important to note that the K-means algorithm was applied here
on the 20 solutions overlapping, as in the left side of Figure 5. First, as a sample,
the number of clusters was set to 2 (case 1), and the result was two clusters of 11
and 9 solutions, as shaded differently (black and grey) in the upper part of Figure 5.
Next, we changed the number of clusters to 3 (case 2) and retrieved three different
groups (shown in black, grey, and light grey), that can be seen in the lower part of
Figure 5. To show their forms, the solutions of both cases were re-oriented, and
distributed along an array of 20 cells, as in the right side of Figure 5.

Figure 1. Workflow of the prototype, representing a combination of several processes. The

major addition to current optimization models is process 3.

Figure 2. Axonometric view of 8 possible options of the grid-based pattern.

Figure 3. From left to right: 3D view, top view, and illuminance mesh of solution 1 and 2.
438 S. YOUSIF AND W. YAN

Figure 4. Octopus interface showing the solution space and two thumbnails of solution 1 and 2,
illustrated in Figure 3.

Figure 5. Above, case 1: two clusters generated from the K-means algorithm, below, case 2:
three clusters.

4. Discussion of Test-Case Outcome

The results of K-means clustering are preliminary, and more experiments are
needed to verify the findings. Often, the results of K-means clustering can be best
represented in a solutions space, which is typically a two or three-dimensional
space, but here it cannot be directly illustrated as the dimensions were ten. The
solutions are analyzed and compared when they overlap (with the grid-center
unit overlapping in all solutions), and this overlap is thought to be reliable for
accurate comparison of the grid-based patterns. It is obvious that more center
points overlapping means similar forms, and leads to cluster their forms together.
There was no mirror, move, or rotation procedures here to recognize more similar
patterns. Comparing the two cases pursued (case 1 and case 2), it can be noted
that the clusters were formed slightly differently. This behavior can be attributed
to the inherent characteristic of K-means that alternates between updating the
assignments of solutions to clusters (Raykov et al. 2016). The outcome of
the Owl-based K-means definition used here was mainly the clusters and their
assigned members. Another expected outcome was the centroid of each cluster,
which has not resulted. Therefore, the representative candidate of each cluster
could not be retrieved in this experiment. Adding the task of obtaining the
centroids to the current algorithm is important as a next step for this work. It is
 CLUSTERING FORMS FOR ENHANCING ARCHITECTURAL DESIGN 439
OPTIMIZATION

important to note that in a typical K-means clustering, the greater homogeneity

within a group, the greater the difference between groups is, leading to more
distinct clustering (Tan 2006, Tan et al. 2013). This characteristic of K-means
is important and advantageous to the form comparison targeted in this study.

5. Conclusions and Future Work

This work is part of an ongoing research that investigates incorporating form
diversity and K-means cluster analysis to create a new architectural design
optimization system. Although a simple test-case was used, the form diversity
algorithm has successfully led to filtering highly diverse forms. The initial
results of K-means clustering are consistent with the characteristics of the typical
K-means method, yet, verification procedures need to be conducted to compare
the results across different scenarios and test-cases. This work aims to contribute
to extend the capabilities of architectural design optimization by creating a method
to be applied to general cases of building forms. Eventually, the method will
be an open-source program, accessible and connected to available modeling and
optimization platforms. In parallel to this work, in a recent publication, we have
investigated the possibility of incorporating architectural aesthetic variables of
modern language as constraints and parameters into initial modeling to enhance
the aesthetic quality of design solutions generated from optimization (Yousif et al.
2018). The aesthetic variables will be incorporated into our system, in addition to
form diversity and K-means algorithms.

References
“LEED BD+C: New Construction | v4 - LEED v4 Daylight. LEED Reference Guide For
Green Building Design And Construction. US Green Building Council (USGBC)” :
2014. Available from <https://www.usgbc.org/credits/new-construction-commercial-inte
riors-schools-new-construction-retail-new-construction-ret-1?view=forum> (accessed 20th
December 2017).
Aish, R., Fisher, A., Joyce, S. and Marsh, A.: 2012, Progress towards multi-criteria design
optimisation using DesignScript with SMART form, robot structural analysis and Ecotect
building performance analysis, Proceedings of ACADIA: Synthetic Digital Ecologies, San
Francisco, CA, 46-56.
Aish, R. and Woodbury, R.: 2005, Multi-level interaction in parametric design, Smart Graphics,
5th International Symposium, Frauenwörth Cloister, Germany, 151-162.
Attia, S., Hamdy, M., O’Brien, W. and Carlucci, S.: 2013, Assessing gaps and needs for
integrating building performance optimization tools in net zero energy buildings design,
Energy and Buildings, 60, 110-124.
Bradner, E. and Davis, M.: 2013, Design creativity: using Pareto analysis and genetic algorithms
to generate and evaluate design alternatives, Proceedings of CHI’13, Paris, France.
Brown, N., Tseranidis, S. and Mueller, C.: 2015, Multi-objective optimization for diversity and
performance in conceptual structural design, Proceedings of the International Association
for Shell and Spatial Structures (IASS).
Caldas, L. and Norford, L.: 2003, Shape generation using pareto genetic algorithms:
integrating conflicting design objectives in low-energy architecture, International journal
of architectural computing, 1(4), 503-515.
Caplan, B.: 2011, Parametric Design’s Greatest Value to Architecture Is to Attain
Eco-sustainability, The Architectural Review, EMAP Architecture.
 440 S. YOUSIF AND W. YAN

Gerber, D., Lin, S.H.E., Pan, B.P. and Solmaz, A.S.: 2012, Design optioneering:
multi-disciplinary design optimization through parameterization, domain integration and
automation of a genetic algorithm, Proceedings of the 2012 Symposium on Simulation for
Architecture and Urban Design.
Hartigan, J. and Wong, M.: 1979, Algorithm AS 136: A k-means clustering algorithm, Journal
of the Royal Statistical Society. Series C (Applied Statistics), 28(1), 100-108.
Jain, A.K.: 2010, Data clustering: 50 years beyond K-means, Pattern recognition letters, 31(8),
651-666.
Jain, A.K. and Dubes, R.C.: 1988, Algorithms for clustering data, Prentice-Hall, Inc..
Kline, M.: 1990, Mathematical Thought From Ancient to Modern Times: Volume 3, OUP USA.
Konak, A., Coit, D.W. and Smith, A.E.: 2006, Multi-objective optimization using genetic
algorithms: A tutorial, Reliability Engineering & System Safety, 91(9), 992-1007.
Konis, K., Gamas, A. and Kensek, K.: 2016, Passive performance and building form: An
optimization framework for early-stage design support, Solar Energy, 125, 161-179.
Michalek, J., Choudhary, R. and Papalambros, P.: 2002, Architectural layout design
optimization, Engineering optimization, 34(5), 461-484.
Radford, A.D. and Gero, J.S.: 1987, Design by optimization in architecture, building, and
construction, John Wiley & Sons, Inc..
Raykov, Y.P., Boukouvalas, A., Baig, F. and Little, M.A.: 2016, What to do when K-means
clustering fails: a simple yet principled alternative algorithm, PloS one, 11(9), e0162259.
Roudsari, M.S., Pak, M. and Smith, A.: 2013, Ladybug: a parametric environmental plugin for
grasshopper to help designers create an environmentally-conscious design, Proceedings of
the 13th International IBPSA Conference, Lyon, France.
Tan, P.N.: 2006, Introduction to data mining, Pearson Education India.
Tan, P.N., Steinbach, M. and Kumar, V. 2013, Data mining cluster analysis: basic concepts and
algorithms, in P.N. Tan, M. Steinbach, A. Karpatne and V. Kumar (eds.), Introduction to
Data Mining, Second Edition., Pearson Education.
Toffolo, A. and Benini, E.: 2003, Genetic diversity as an objective in multi-objective
evolutionary algorithms, Evolutionary Computation, 11, 151-167.
Vierlinger, R.: 2013, Multi objective design interface, Master’s Thesis, Technischen Universität
Wien.
Wang, W., Rivard, H. and Zmeureanu, R.: 2006, Floor shape optimization for green building
design, Advanced Engineering Informatics, 20(4), 363-378.
Welle, B., Fischer, M., Haymaker, J. and Bazjanac, V.: 2012, CAD-centric attribution
methodology for multidisciplinary optimization (CAMMO): enabling designers to
efficiently formulate and evaluate large design spaces, Center for Integrated Facility
Engineering: CIFE, Technical Report, TR195.
Wurzer, G., Pont, U., Lorenz, W. and Mahdavi, A.: 2016, Coupling Building Morphology
Optimization and Energy Efficicency - A Proof of Concept, Proceedings of BauSim 2016,
Dresden, Germany.
Yan, W., Asl, M.R., Su, Z. and Altabtabai, J.: 2015, Towards Multi-Objective Optimization
for Sustainable Buildings with Both Quantifiable and Non-Quantifiable Design Objectives,
Proceedings of Sustainable Human–Building Ecosystems, 223-230.
Yi, H., Srinivasan, R.S. and Braham, W.W.: 2015, An integrated energy-emergy approach to
building form optimization: Use of EnergyPlus, emergy analysis and Taguchi-regression
method, Building and Environment, 84, 89-104.
Yousif, S., Claytom, M. and Yan, W.: 2018, Towards Integrating Aesthetic Variables in
Architectural Design Optimization, Proceedings of the 106th ACSA Annual Meeting: The
Ethical Imperative, Denver, CO.
Yousif, S., Yan, W. and Culp, C.: 2017, Incorporating Form Diversity into Architectural Design
Optimization, Proceedings of the 37th Annual Conference of the Association for Computer
Aided Design in Architecture (ACADIA): DISCIPLINES & DISRUPTION, MIT, Cambridge,
MA, 640-649.

View publication stats