Improved strength prediction of cemented paste backfill using a novel model based on adaptive neuro fuzzy inference system and artificial bee colony

Abstract

The cemented paste backfill (CPB) technology has matured as a promising way for tailings recycling in the mining industry. Nevertheless, the current CPB design requires a large number of lab experiments to determine the unconfined compressive strength (UCS) of CPB. The utilisation of artificial intelligence (AI) prediction to reduce the lab experiments has been attempted without reaching its full potential. In this study, a hybrid model based on adaptive neuro fuzzy inference system (ANFIS) and artificial bee colony (ABC) was used for performance improvement. The ANFIS was used to learn the relationship between inputs and UCS while the ABC algorithm was used to tune the parameters of the initial ANFIS. The convergence of the prediction performance was tested using Monte Carlo simulations. A comparison between this study and previous studies was conducted and a sensitivity analysis was performed to investigate the importance of input variables. The results show that the ABC algorithm was efficient in tunning parameters of the ANFIS model. The representative ANFIS-ABC model yielded an R2 of 0.967 on the training part and an R2 of 0.976 on the testing part, indicating an excellent prediction. 310 Monte Carlo simulations were needed before a stable performance was achieved for all quality assessment criteria. The proposed hybrid model outperformed AI models in the literature (R2 was increased from 0.83/0.958/0.86 to 0.976 on the testing set). Solid content, cement-tailings ratio and curing time were found to be the most significant input parameters for the UCS of CPB.