Civil Engineering Reference
In-Depth Information
Fig. 7 Basic genetic
algorithm pseudo-code
Begin
INITIALIZE
population with random candidate solutions;
EVALUATE
each candidate;
REPEAT
1
SELECT
parents;
2
RECOMBINE
parents;
3
MUTATE
the resulting offspring;
4 EVALUATE new candidates;
5 SELECT individuals for the next generation
UNTIL (TERMINATION CONDITION is satisfied)
END
applications, GA are frequently used for the optimization of building thermal
system design (Wright et al.
2002
), HVAC controls (Huang and Lam
1997
;Lu
et al.
2005
), and chiller energy costs (Chow et al.
2002
).
The pseudo-algorithm of GA is displayed in Fig.
7
and can be described with
the following steps:
• First, a random population is created, where each individual represents a solu-
tion using some encoding scheme (for instance, binary).
• At each generation, couples of individuals (parents) produce new individuals by
gene crossover and mutation (offspring).
• At the end of each generation, the candidate solutions to be included in the next
generation are evaluated using a fitness evaluation function.
• The last two steps operate until the termination condition is met (generally based
on the number of generations or on the stagnancy of population fitness).
As a gradient-free method, GA is able to deal with nonlinear functions and to
find global optima without being trapped in local ones. Furthermore, it can handle
real, discrete, or even discontinuous variables and can be applied to noisy
objective functions (Wright et al.
2002
; Huang and Lam
1997
).
A main drawback of GA is the high burden whenever it is necessary to make a
high number of calls to an evaluation function involving a high computational
cost. In building applications, these evaluations are generally estimated by an
external simulation program such as CFD or other simulation packages. If accurate
results are required, each evaluation can be time consuming, and thus the complete
computational process becomes extremely unattractive (Magnier
2008
). For
instance, for the two-objective optimization of building floor shape, Wang et al.
(
2006
) used an evaluation tool where each evaluation took 24 s (CPU time). In that
case, the total optimization time, which is mainly due to evaluations, was 68 h.
Using simulation software where each evaluation would take several minutes, a
similar optimization would result in a total optimization time of several months.
This shortcoming should be dealt with before being able to take full advantage of
GA in building energy efficiency problems.
Genetic algorithm integrating neural network (GAINN) is one of the solutions
to the above-mentioned problem. The main idea of GAINN is to benefit from the
