Neural network and particle swarm optimization for predicting the unconfined compressive strength of cemented paste backfill

Abstract

Cemented paste backfill (CPB) has been widely used to prevent and mitigate hazards produced during the excavation of underground stopes. In practice, the strength of CPB is often an essential parameter for successful stope design. We propose an intelligent technique in this study for predicting the unconfined compressive strength (UCS) of CPB. This intelligent technique is a combination of the artificial neural network (ANN) and particle swarm optimization (PSO). The ANN was used for non-linear relationships modelling and PSO was used for the ANN architecture-tuning. Inputs of the ANN were selected to be the tailings type, the cement-tailings ratio, the solids content, and the curing time. A total of 396 CPB specimens under different combination of influencing variables were tested for the preparation of the dataset. The results showed that PSO was efficient for the ANN architecture-tuning. Also, comparison of the predicted UCS values with experimental values showed that the optimum ANN model was very accurate at predicting CPB strength.