Risk assessment of water inrush from coal floor based on enhanced samples with class distribution

Area of Science:

• Mining Engineering
• Geological Engineering
• Data Science

Background:

• Water inrush from coal floors poses a significant risk to mining safety.
• Limited and random on-site data hinder accurate prediction model development.
• Existing models struggle with small sample sizes and generalizability.

Purpose of the Study:

• To develop a robust method for enhancing virtual sample databases for water inrush risk assessment.
• To improve the accuracy and generalizability of water inrush prediction models.
• To support safe and efficient coal mining above Ordovician limestone-confined water.

Main Methods:

• Proposed a virtual sample enhancement method using class distribution mega-trend diffusion technology (CDMTD).
• Introduced constraints on the class distribution of influencing factors for virtual sample generation.
• Developed a prediction model using a coupled algorithm: PCA-CDMTD-SaDE-ELM.
• Applied the model to evaluate water inrush risk in a specific mine working face.

Main Results:

• The CDMTD method effectively enhanced the measured database and mitigated small sample size issues.
• The PCA-CDMTD-SaDE-ELM model demonstrated superior prediction performance compared to other optimization models.
• Achieved a significant error reduction of 42.95-51.27% with results biased towards safety.

Conclusions:

• The proposed CDMTD-based virtual sample enhancement is effective for improving water inrush risk prediction.
• The coupled PCA-CDMTD-SaDE-ELM model offers a reliable tool for assessing water inrush risk.
• The findings contribute to the safe and efficient exploitation of coal resources above confined aquifers.