This work is intended to optimize the thermal behavior of a building, by means of the proper selection of some of its design parameters, such as thickness, thermal properties and reflectivity of walls and roof, using a tropical building as an example. The solution methodology used in this work involved several stages. First, a data set was built using the Latin Hypercube sampling method. Then, for every sample of the data set, the maximum daily value of the mean temperature in the inner space of the building was determined, using a computationally expensive numerical simulator. These results, along with the data set, were used to train and validate a three-layer feed forward neural network, which was used as a surrogate model for two global optimization algorithms. This approach leads to a significant reduction of the time requirements of the design cycle. The results show how a proper selection of design parameters can improve the thermal behavior of the building. The improvements obtained were in the range between 3 and 11 Celsius degrees.
[1]
Lester Ingber,et al.
Simulated annealing: Practice versus theory
,
1993
.
[2]
Richard J. Beckman,et al.
A Comparison of Three Methods for Selecting Values of Input Variables in the Analysis of Output From a Computer Code
,
2000,
Technometrics.
[3]
PhD DEng D. T. Pham BE,et al.
Intelligent Optimisation Techniques
,
2000,
Springer London.
[4]
David E. Goldberg,et al.
Genetic Algorithms in Search Optimization and Machine Learning
,
1988
.
[5]
Ali Zilouchian,et al.
FUNDAMENTALS OF NEURAL NETWORKS
,
2001
.
[6]
D. E. Goldberg,et al.
Genetic Algorithms in Search
,
1989
.