Building Performance Optimization of Net Zero-Energy Buildings - Modeling, Design, and Optimization of Net-Zero Energy Buildings - page 255

fitness of a population.” In typical implementations, design variables are
represented using binary or discrete formats. Genetic algorithms are a
well-studied group within the broader metaheuristic family. Wang, Rivard,
andZmeureanu(2006)usedaGAtoperformamulti-objectiveoptimization
using lifecycle cost and exergy on a green building with a polygonal-shaped
floor plan. Caldas (2008) used a GA to simultaneously optimize building
geometry, energy efficiency, and visual comfort. Many modifications exist
combining the best elements of other search strategies from the
evolutionary algorithm family, such as Differential Evolution (DE) (Price,
Storn, and Lampinen, 2005), Evolutionary Strategies (ES) (Eiben and
Smith, 2003), and Genetic Programming (GP) (Poli, Langdon, and McPhee,
2008). Literature commonly refers to a modified GA by their more general
family name, EA, to avoid confusion. EAs have been scaled to building
optimizationproblemswithmanydesignvariables.Forexample,Kampfand
Robinson (2010) optimized the layout of a buildings cluster to maximize
available solar radiation, while considering design parameters, such as
insulation in ceilings and walls, window types and areas, infiltration, and
thermal mass. A benefit of EAs is the flexibility to include subspecialized
search strategies. For example, multi-island EAs allow for the population
in one generation to be divided into subpopulations, or islands, where
specialized subpopulation searches can be performed. This approach is
useful to deconstruct large optimization problems into smaller, more
solvable problems. Ooka and Komamura (2009) utilized a multi-island EA
to design, schedule, and control an HVAC system for a hospital in Japan.
A particle swarm optimization (PSO) is fundamentally different from
evolutionary cycles found in EAs (Eberhart and Kennedy, 1995). Instead
of forming a new population of individuals at each iteration, the existing
population is allowed to gravitate toward other more fit individuals, or
particles, in the population. Particles are updated using the best local and
global particles in the swarm. Representations are vectors of continuous
design variables, although binary and discrete representations can also be
used (Kennedy and Eberhart, 1997). PSO competes favorably with other
optimization algorithms. For example, Elbeltagi, Hegazy, and Grierson
(2005) compared five evolutionary-based algorithms and found that PSO
outperforms the other algorithms for a discrete design problem, with regard
to reproducibility of optimal solutions and ability to scale with increasing
problem sizes. PSOs are the primary population-based search approach
usedintheGenericOptimization Program(GenOpt)(Wetter,2001).Hasan,