A Comparative Study of Six Hybrid Prediction Models for Uniaxial Compressive Strength of Rock Based on Swarm Intelligence Optimization Algorithms

Abstract

Uniaxial compressive strength (UCS) is a significant parameter in mining engineering and rock engineering. The laboratory rock test is time-consuming and economically costly. Therefore, developing a reliable and accurate UCS prediction model through easily obtained rock parameters is a good way. In this paper, we set five input parameters and compare six hybrid models based on BP neural network and six swarm intelligence optimization algorithms–bird swarm algorithm (BSA), grey wolf optimization (GWO), whale optimization algorithm (WOA), seagull optimization algorithm (SOA), lion swarm optimization (LSO), firefly algorithm (FA) with the accuracy of two single models without optimization–BP neural network and random forest algorithm. Finally, the above eight models were evaluated and compared by root mean square error (RMSE), mean absolute percentage error (MAPE), coefficient of determination (R2), and a10 index to obtain the most suitable prediction model. It is indicated that the best prediction model is the FA-BP model, with a RMSE value of 4.883, a MAPE value of 0.063, and a R2 of 0.985, and an a10 index of 0.967. Furthermore, the normalized mutual information sensitivity analysis shows that point load strength is the most effective parameters on the UCS, respectively.