Open AccessModeling ambient air temperature time series using neural networks
Author(s) -
Mihalakakou G.,
Santamouris M.,
Asimakopoulos D.
Publication year - 1998
Publication title -
journal of geophysical research: atmospheres
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.67
H-Index - 298
eISSN - 2156-2202
pISSN - 0148-0227
DOI - 10.1029/98jd02002
Subject(s) - autoregressive model , artificial neural network , series (stratigraphy) , backpropagation , air temperature , time series , computer science , meteorology , environmental science , statistics , mathematics , machine learning , geology , geography , paleontology
A neural network approach is used in this study to analyze and model the ambient air temperature time series. The model's predictions can be very useful in Meteorology, atmospheric sciences, and in energy applications such as the control of conventional and passive cooling systems in order to achieve thermal comfort inside buildings. The future hourly values of ambient temperature for several years were predicted, using multiple‐layer backpropagation networks, by extracting knowledge from its past values. The results were tested using various sets of nontraining measurements, and it was found that predicted values correspond well to the actual values. Furthermore, “multilag” output predictions were performed using the predicted values to the input database in order to model future air temperature values with sufficient accuracy. The neural network model predictions were compared with the results of an autoregressive linear model. It was found that the neural network model makes significantly better predictions than the autoregressive model.