Prediction of UCS based on multivariate adaptive regression splines and the BP neural network

Uniaxial compressive strength is one of the most important parameters for rock characterization in geotechnical engineering. Based on 45 rock samples with different diameters (54 mm, 48 mm, and 42 mm) obtained from sandstone, limestone and concrete, the relationships among the uniaxial compressive strength and point load strength, longitudinal wave velocity, and rock sample diameter are studied. The 45 groups of data are divided into a training set (36 groups) and test set (9 groups). Regression analysis was carried out using Multivariate Adaptive Regression Splines (MARS), the BP neural network and multiple linear regression, and finally, the best regression model was obtained. The applicability of the BP neural network and MARS model is better than that of the multiple linear regression model. Among these models, the BP neural network has the best fitting effect and the R² values of the training set and test set of the BP neural network reach 0.986 and 0.958, respectively. In addition, the R² values of the training set and test set of the MARS model reach 0.916 and 0.846, respectively. The results show that the BP neural network has high accuracy and practicability in uniaxial compressive strength prediction.