Artificial intelligence models for predicting the compressive strength of cemented backfill with waste rock tailings

DOI registering

Abstract

Cemented backfilling using waste rock and tailings is a promising backfilling method in underground mining. The compressive strength of a cemented backfill influences the stability of the surrounding rock and backfill body, the backfill order, and the stress-changing process. This study presents two hybrid artificial neural network (ANN) models to predict the compressive strength of cemented backfill using the wild horse optimiser (WHO) and the coronavirus herd immunity optimiser (CHIO). The dataset used in this study included 140 records, and seven variables were considered input parameters: weight concentration, bulk concentration, cement, aggregate, water, water-cement ratio, and cement-sand ratio. Model performance was evaluated using five common metrics. The results demonstrated that the WHO-ANN model exhibited robust predictive capabilities, achieving MAE = 0,305, RMSE = 0,388, R-squared = 0,928, and NSE = 0,921. These indicators reflect high accuracy and an excellent correlation between predicted and observed values. The sensitivity analysis indicated that the cement-sand ratio and cement were important input variables. Therefore, it was concluded that the hybrid ANN models were robust and efficacious. This study is anticipated to enable the rapid estimation of the compressive strength of cemented backfill, thereby reducing costs and saving time.