Construction and Application of VMD-Transformer-LSTM-SVM Time Series Prediction Model for Water Inflow in Coal Seam Roofs

Xuan LIU , Ya-dong JI , Kai-peng ZHU , Chun-hu ZHAO , Kai LI , Pan-pan LI , Chen-han YUAN , Rong-long MU , Peng-zhen YAN

Science Technology and Engineering ›› 2026, Vol. 26 ›› Issue (18) : 7747 -7758.

PDF
Science Technology and Engineering ›› 2026, Vol. 26 ›› Issue (18) : 7747-7758 DOI: 10.12404/j.issn.1671-1815.2504573
research-article
Construction and Application of VMD-Transformer-LSTM-SVM Time Series Prediction Model for Water Inflow in Coal Seam Roofs
Author information +
History +
PDF

Abstract

Mine water hazards in coal mining have currently become the second major “killer” in China’s coal mining. Accurately grasping the variation law of mine water inrush and precisely predicting the mine water inrush volume are of great significance for the allocation of mine resources and safe production. In order to construct a mine water inflow prediction model applicable to the impact of the massive and thick sandstone water hazard in western China, taking the Binchang mining area, which is most severely threatened by this water hazard, as the research object, based on the water inflow data of the 3407 working face of Gaojiabao certain mine. A water inflow prediction model for mine working faces based on VMD (variational mode decomposition)-LSTM(transformer-long short-term memory)-SVM(support vector machine) was proposed. The original water inflow time series data were first decomposed into sub-sequences of each mode through VMD. Features were then extracted for each modal component using the self-attention mechanism of the Transformer architecture. Finally, these features were fed into the LSTM-SVM combined model to complete the subsequent prediction of water inflow. The model prediction results show that the VMD-Transformer-LSTM-SVM coupling model has the smallest error compared with the other models: EMAE、EMAPE and ERMSE are 47.76 m3/h, 3.33% and 57.34 m3/h respectively, with a maximum R2 of 0.96. Moreover, compared with the prediction results of each model that have not undergone VMD, they also show higher robustness and accuracy. The research finds that the VMD-Transformer-LSTM-SVM coupling model shows relatively obvious advantages in the prediction of water inflow in the working face of this mine. This model not only enriches the prediction methods of water inflow in western mines, but also provides a new method for predicting mine water inflow for mines with geological conditions similar to the study mine. The research results have certain practical significance for the prediction of mine water inflow in mining areas and the prevention and control of mine water hazards.

Graphical abstract

Keywords

Binchang Mining Area / water inflow prediction / VMD / Transformer / LSTM-SVM

Cite this article

Download citation ▾
Xuan LIU, Ya-dong JI, Kai-peng ZHU, Chun-hu ZHAO, Kai LI, Pan-pan LI, Chen-han YUAN, Rong-long MU, Peng-zhen YAN. Construction and Application of VMD-Transformer-LSTM-SVM Time Series Prediction Model for Water Inflow in Coal Seam Roofs. Science Technology and Engineering, 2026, 26(18): 7747-7758 DOI:10.12404/j.issn.1671-1815.2504573

登录浏览全文

4963

注册一个新账户 忘记密码

References

Funding

PDF (13416KB)

29

Accesses

0

Citation

Detail

Sections
Recommended

/

〈 〉