A prediction model for blasted block size grouping based on HC and RF-GA-BP neural network

Abstract

Blast block prediction is a complex non-stationary, nonlinear problem, the contribution of factors affecting results varies for different external conditions. Studies in a single environment are not universally applicable, the establishment of a blasted block size prediction model with fusion of multiple algorithms and reliable prediction results is the most urgent problem to be solved. In this study, a method is proposed that applies to different regions and rock conditions. To achieve the grouping prediction of blasted block size, this study firstly used hierarchical clustering to cluster the data in different areas, then used the random forest to establish the data grouping discriminant model based on the grouping results, and used back propagation neural network with genetic algorithm (GA-BP) to establish the blasted block size prediction model for each group of the blasted block size data separately. The results of the study show that (1) the block size data with different properties can be grouped according to the elastic modulus; (2) the grouping discriminant model established can correctly group the data; (3) compared with the GA-BP neural network prediction model without grouping, this model has a higher coefficient of determination (${R}^{2}=0.982$ ) and smaller root mean square error ($\mathrm{RMSE}=0.2$) and mean relative error ($\mathrm{MRE}=4.857$ ), which verifies the correctness of grouping prediction according to the elastic modulus; (4) by comparing with multiple regression, least squares support vector machines (LSSVM), and BP neural networks, the model outperforms the other models in ${R}^{2}$$\mathrm{RMSE}$, and $\mathrm{MRE}$, and the prediction results are more accurate.