Improving pressure drops estimation of fresh cemented paste backfill slurry using a hybrid machine learning method

Estimation of pressure drops of fresh cemented paste backfill slurry is a novel idea with great potentials. This paper presented a hybrid machine learning (ML) method for improved pressure drops estimation using a combination of artificial neural network and differential evolution. A comprehensive parametric study was conducted on training dataset size (N-size), ML methods, and Monte Carlo random sampling. Moreover, dependent analysis of pressure drops to each influencing variable was performed. The results indicate that 300 Monte Carlo realizations were sufficient for the converged and reliable results. The optimum N-size was determined to be 70%, and the proposed hybrid method outperformed six individual ML methods. The estimation performance has been significantly improved compared to the methods used in the literature (R-2 increased from 0.83 to 0.95 on the testing dataset). Solids content, inlet velocity, SiO2, CaO, and Fe2O3 were determined to be the most significant variables for pressure drops.

Automated summary

Estimation of pressure drops for fresh cemented paste backfill slurry using a hybrid machine learning method combining artificial neural network and differential evolution has shown significant improvement in estimation performance. The most influential variables on pressure drops were found to be solids content, inlet velocity, SiO2, CaO, and Fe2O3.