Estimation of the Daily Heating and Cooling Loads Using Artificial Neural Networks
Date Issued
September 2001
Abstract
The objective of this work is to use Artificial Neural Networks for the estimation of the daily
heating and cooling loads. The daily loads of nine different building structures have been
estimated using the TRNSYS program and a typical meteorological year of Cyprus. This set
of data has been used to train a neural network. For each day of the year the maximum and
minimum loads were obtained from which heating or cooling loads can be determined. All the
buildings considered, had the same areas but different structural characteristics. Single and
double walls have been considered as well as a number of different roof insulations. A multislab
feedforward architecture having 3 hidden slabs has been employed. Each hidden slab
comprised of 36 neurons. For the training data set the R2-values obtained were 0.9896 and
0.9918 for the maximum and minimum loads respectively. The method was validated by
using actual (modeled) data for one building, for all days of the year, which the network has
not seen before. The R2-values obtained in this case are 0.9885 and 0.9905 for the two types
of loads respectively. The results indicate that the proposed method can be used for the
required predictions for buildings of different constructions. At present the method was used
primarily to investigate its suitability for this kind of predictions.
heating and cooling loads. The daily loads of nine different building structures have been
estimated using the TRNSYS program and a typical meteorological year of Cyprus. This set
of data has been used to train a neural network. For each day of the year the maximum and
minimum loads were obtained from which heating or cooling loads can be determined. All the
buildings considered, had the same areas but different structural characteristics. Single and
double walls have been considered as well as a number of different roof insulations. A multislab
feedforward architecture having 3 hidden slabs has been employed. Each hidden slab
comprised of 36 neurons. For the training data set the R2-values obtained were 0.9896 and
0.9918 for the maximum and minimum loads respectively. The method was validated by
using actual (modeled) data for one building, for all days of the year, which the network has
not seen before. The R2-values obtained in this case are 0.9885 and 0.9905 for the two types
of loads respectively. The results indicate that the proposed method can be used for the
required predictions for buildings of different constructions. At present the method was used
primarily to investigate its suitability for this kind of predictions.
File(s)![Thumbnail Image]()
Name
C41-CLIMA2001.pdf
Size
309.13 KB
Format
Adobe PDF
Checksum (MD5)
4612d1aca0bbb6ff2789aedd374c8665