Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Spatio-temporal interpretable neural network for solar irradiation prediction using transformer
https://orcid.org/0000-0002-3526-5856
https://orcid.org/0000-0001-8839-9834
Abstract Deep learning models have been increasingly applied in the field of solar radiation prediction. However, the characteristics of a deep learning black box model restrict its application in practical scenarios such as model predictive control. Because energy system controllers may be unable to make final decisions based solely on the predictions of a black-box model. This study considers both the temporal and spatial dependencies of solar radiation predictions through unfolding sequences and applying a transformer model As the results indicate, the transformer model used can improve the mean absolute percent error by approximately 20.9% and the mean squared error by 14.3% compared to the baseline recurrent neural network model. At the same time, detailed case studies show that the transformer model heavily considers humidity and temperature when predicting the more significant outcomes Finally, the detailed results of a one-step analysis prove that the change in weight of the transformer model is related to the change in outdoor weather conditions.
Spatio-temporal interpretable neural network for solar irradiation prediction using transformer
https://orcid.org/0000-0002-3526-5856
https://orcid.org/0000-0001-8839-9834
Abstract Deep learning models have been increasingly applied in the field of solar radiation prediction. However, the characteristics of a deep learning black box model restrict its application in practical scenarios such as model predictive control. Because energy system controllers may be unable to make final decisions based solely on the predictions of a black-box model. This study considers both the temporal and spatial dependencies of solar radiation predictions through unfolding sequences and applying a transformer model As the results indicate, the transformer model used can improve the mean absolute percent error by approximately 20.9% and the mean squared error by 14.3% compared to the baseline recurrent neural network model. At the same time, detailed case studies show that the transformer model heavily considers humidity and temperature when predicting the more significant outcomes Finally, the detailed results of a one-step analysis prove that the change in weight of the transformer model is related to the change in outdoor weather conditions.
Spatio-temporal interpretable neural network for solar irradiation prediction using transformer
https://orcid.org/0000-0002-3526-5856
https://orcid.org/0000-0001-8839-9834
Abstract Deep learning models have been increasingly applied in the field of solar radiation prediction. However, the characteristics of a deep learning black box model restrict its application in practical scenarios such as model predictive control. Because energy system controllers may be unable to make final decisions based solely on the predictions of a black-box model. This study considers both the temporal and spatial dependencies of solar radiation predictions through unfolding sequences and applying a transformer model As the results indicate, the transformer model used can improve the mean absolute percent error by approximately 20.9% and the mean squared error by 14.3% compared to the baseline recurrent neural network model. At the same time, detailed case studies show that the transformer model heavily considers humidity and temperature when predicting the more significant outcomes Finally, the detailed results of a one-step analysis prove that the change in weight of the transformer model is related to the change in outdoor weather conditions.
Spatio-temporal interpretable neural network for solar irradiation prediction using transformer
Gao, Yuan (Autor:in) / Miyata, Shohei (Autor:in) / Matsunami, Yuki (Autor:in) / Akashi, Yasunori (Autor:in)
Energy and Buildings ; 297
13.08.2023
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
A Spatio-Temporal Graph Convolutional Network for Air Quality Prediction
| DOAJ | 2023
CT-Net: an interpretable CNN-Transformer fusion network for fNIRS classification
| Springer Verlag | 2024
Spatio-Temporal Wind Speed Prediction Based on Improved Residual Shrinkage Network
| DOAJ | 2023