Prediction of earth-air activity in the extremely arid regions based on the CNN-LSTM-Attention model
Measurement Science and Technology · 10.04.2026 · 37 (15) · 155801
DOI: 10.1088/1361-6501/ae5cba ↗Kurz gefasst
Fachartikel in der Zeitschrift Measurement Science and Technology, online erschienen am 17.04.2026. Verfasst von Fei Li und 2 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben. Gefördert von: Gansu Provincial Science and Technology Program, National Natural Science Foundation of China, Science and Technology Support Program of Dunhuang City.
Abstract (Originalsprache)
The deterioration process of the wall paintings in Mogao Grottoes in Dunhuang is governed by the mechanism of water-salt activity caused by earth-air activity, which driven by the fluctuations of atmospheric pressure (AP). However, there is obvious hysteresis effect in the process of earth-air rising and falling, which will cause the curve of earth-air activity to be ahead of the AP that triggers its variations, forming a ‘paradox’ of causal inversion (This paradox only represents an apparent temporal characteristic and does not involve a substantive reversal of causality.). This study uses the cross-correlation analysis method to systematically analyze the ‘advanced phenomenon’ between earth-air activity and AP variations, and uses the CNN-LSTM-Attention to achieve accurate prediction of earth-air activity on the minute scale. The results show that the response of earth-air flow (EF) in the main chamber of Cave 108 to the variation of AP shows an advanced phase of about 21.833 h, which is significantly longer than that of the drilling well of 12.167 h. The difference in hysteresis time reveals the regulation effect of geological three-dimensional pore structure on AP conduction efficiency, and further confirms that there is a significant spatial hysteresis difference in earth-air activity. The hysteresis of AP conduction in space leads to the prepositioning of the EF curve on the time scale. Notably, the dynamic changes in the concentrations of gases such as CO 2 and 222 Rn in the cave depend not only on the environmental factors inside the cave but also, more importantly, on the historical changes of these factors. On the minute scale, the earth-air prediction model achieves high-precision predictions of earth-air CO₂ concentration ( R 2 = 0.987), Rn concentration ( R 2 = 0.904) and EF ( R 2 = 0.618). At the same time, it also provides a predictive paradigm for the micro-environment of cave heritage.
Förderung
- Gansu Provincial Science and Technology Program
- National Natural Science Foundation of China
- Science and Technology Support Program of Dunhuang City
Zitieren
APA
Li, F., Li, H., & Wang, S. (2026). Prediction of earth-air activity in the extremely arid regions based on the CNN-LSTM-Attention model. Measurement Science and Technology, 37(15), 155801. https://doi.org/10.1088/1361-6501/ae5cba
Vancouver
Li F, Li H, Wang S. Prediction of earth-air activity in the extremely arid regions based on the CNN-LSTM-Attention model. Measurement Science and Technology. 2026;37(15):155801. doi:10.1088/1361-6501/ae5cba
BibTeX
@article{Li2026,
author = {Li, Fei and Li, Hongshou and Wang, Shunren},
title = {Prediction of earth-air activity in the extremely arid regions based on the CNN-LSTM-Attention model},
journal = {Measurement Science and Technology},
year = {2026},
volume = {37},
number = {15},
pages = {155801},
doi = {10.1088/1361-6501/ae5cba}
}
EUVIMED