Strength prediction and application of cemented paste backfill based on machine learning and strength correction

Area of Science:

• Mining Engineering
• Materials Science
• Artificial Intelligence

Background:

• Cemented paste backfill (CPB) is crucial in global mining for ground support.
• CPB strength is influenced by factors like slurry concentration and cement content.
• Accurate prediction of unconfined compressive strength (UCS) is vital for mine safety.

Purpose of the Study:

• To investigate the unconfined compressive strength (UCS) of cemented paste backfill (CPB).
• To develop and optimize an intelligent prediction model for CPB UCS.
• To provide a reliable method for predicting CPB strength in actual mining engineering.

Main Methods:

• Laboratory experiments were conducted using 180 sets of UCS data.
• Machine learning models including BPNN, RBFNN, GRNN, and LSTM were trained for UCS prediction.
• The Long Short-Term Memory (LSTM) model was optimized using Grey Wolf Optimizer (GWO) for enhanced performance.
• A correction coefficient (k) was introduced to bridge laboratory and engineering predictions.

Main Results:

• LSTM demonstrated optimal prediction performance among the tested neural networks.
• The GWO-LSTM model effectively captured the non-linear relationships influencing CPB UCS.
• The optimized GWO-LSTM model achieved high accuracy (VAF = 98.2847%) with low error (RMSE = 0.0204).
• The combined GWO-LSTM and correction coefficient (k) successfully predicted CPB strength in 153 diverse engineering cases.

Conclusions:

• The GWO-LSTM model offers a robust and intelligent approach for predicting CPB UCS.
• The integration of a correction coefficient enhances the model's applicability to real-world engineering scenarios.
• This study provides valuable guidance and an advanced intelligent method for safe mining practices.