|
| Time Series Prediction Model for Internal and External Temperature of Confined Space |
| Received:February 15, 2023 Revised:May 10, 2023 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7643/issn.1672-9242.2023.11.021 |
| KeyWord:confined space internal and external temperature time series prediction physical mechanism multi-variable time series long and short term memory network |
| Author | Institution |
| ZHOU Jun-yan |
Southwest Institute of Technology and Engineering, Chongqing , China |
| WANG Jing-cheng |
Southwest Institute of Technology and Engineering, Chongqing , China |
| SHU Chang |
Southwest Institute of Technology and Engineering, Chongqing , China |
| HUANG Lun |
Southwest Institute of Technology and Engineering, Chongqing , China |
| ZHANG Zhi-hao |
Southwest Institute of Technology and Engineering, Chongqing , China |
| ZHANG Kai |
Dunhuang Atmospheric Material Corrosion Field National Observation and Research Station, Gansu Dunhuang , China |
|
| Hits: |
| Download times: |
| Abstract: |
| ed as a multi-variable time series prediction problem, and the correlation and dependence analysis among variables were the key difficulties. By referring to the ideas of feature fusion, attention mechanism and multi-task model, combined with the physical mechanism and data characteristics, and based on the basic network unit of long and short term memory network, the internal and external temperature time series prediction model of confined space was constructed. The data of a certain type of confined space was collected in Wanning, Dunhuang and Mohe, and different model experiments were carried out based on the data sets of the three places. The multi-variable model had better performance than the univariable model, and the attention mechanism did not significantly improve the performance of this scenario. The model structure design combined with the physical mechanism fully considered the correlation and dependence between variables, which could significantly improve the prediction accuracy. The multi-variable time series prediction model with double input and double output had the highest accuracy and the most stable robustness. It was a relatively optimal model for the prediction of internal and external temperature time series in confined space. The research conclusions can guide the modeling of other environmental characteristics in confined space, and the research ideas can provide references for the correlation and dependency analysis among variables in other multi-variable sequential modeling problems. |
| Close |
|
|
|