Short-Term Mine Water Inflow Forecasting at Sandao Coal Mine Using an Adaptive Weighted Ensemble Model

Shaokai Hu; Xuelei Zhang

doi:10.6919/ICJE.202609_12(9).0010

Authors

Shaokai Hu North China Institute of Science and Technology, Hebei State Key Laboratory of Mine Disaster Prevention, Beijing 101601, China
Xuelei Zhang Shaanxi Changwu Tingnan Coal Industry Co., Ltd., Xianyang, Shaanxi 712000, China

DOI:

https://doi.org/10.6919/ICJE.202609_12(9).0010

Keywords:

Mine Water Inflow; 10-day Interval Forecasting; Adaptive Weighting; Rolling Forecast; Small Sample.

Abstract

The water inflow in mines exhibits dynamic characteristics influenced by multiple coupled factors, and a single predictive model often cannot meet the requirements for precise forecasting. This study proposes an adaptive weighted ensemble for rolling forecasts at 10-day intervals. A continuous series of 117 observations from January 2022 to March 2025 was extracted from the water inflow ledgers of Sandao Coal Mine. Persistence, a three-point damped trend model, and a seasonal-naive model were used as complementary base learners. Their weights were updated from the absolute errors over the latest 18 periods. Data from 2022-2023 were used for model construction and parameter selection, 2024 was retained as an independent test set, and the first quarter of 2025 served as an external validation set. On the 2024 test set, the ensemble achieved an MAE of 10.563 m³/h, an RMSE of 15.881 m³/h, a MAPE of 10.690%, and an R² of 0.653. Relative to persistence, MAE, RMSE, and MAPE decreased by 4.89%, 9.45%, and 6.40%, respectively. The external-validation MAPE was 9.386%. Forecasts for the first, middle, and last 10-day periods of April 2025 were 73.4, 74.9, and 76.0 m³/h, respectively. The proposed method is computationally light, traceable, and suitable for short-term mine water inflow forecasting when monitoring samples are limited.

Downloads

Download data is not yet available.

References

[1] Yin, S. X., Xu, B., Yin, H. C., & et al. (2023). Basic framework and connotation of mine water prevention and control. Coal Science and Technology, 51(7), 24–35. (in Chinese)

[2] Zeng, Y. F., Yu, C., Wu, Q., & et al. (2024). Three-zone management theory for coal-mine water prevention and control and its significance. Journal of China Coal Society, 49(8), 3605–3618. (in Chinese)

[3] Qin, Z. J. (2023). Application and evaluation of the analogy method for mine water inflow prediction. China Metal Bulletin, (11), 222–224. (in Chinese)

[4] Yang, S. L., Lian, H. Q., Xu, B., & et al. (2023). Application of robust deep learning models to predict mine water inflow: Implication for groundwater environment management. Science of the Total Environment, 871, 162056.

[5] Lian, H. Q., Li, Q. X., Wang, R., & et al. (2024). Mine water inflow prediction using a deep-learning LSTM-GRU ensemble model. Safety in Coal Mines, 55(9), 166–172. (in Chinese)

[6] Ding, Y. Y., Yin, S. X., Lian, H. Q., & et al. (2024). Multi-factor coal-face water inflow prediction based on an SSA-CG-Attention model. Coal Geology & Exploration, 52(4), 111–119. (in Chinese)

[7] Bates, J. M., & Granger, C. W. J. (1969). The combination of forecasts. Operational Research Quarterly, 20(4), 451–468.

[8] Hoerl, A. E., & Kennard, R. W. (1970). Ridge regression: Biased estimation for nonorthogonal problems. Technometrics, 12(1), 55–67.

[9] Hyndman, R. J., & Athanasopoulos, G. (2021). Forecasting: Principles and practice (3rd ed.). OTexts.

[10] Zhou, R., Wei, Z. Y., Zhang, Y. B., & et al. (2018). Time series prediction of tomato yield based on LSTM recurrent neural network. Water Saving Irrigation, (8), 66–70.