A comparative application of the Buckingham π (pi) theorem, white-box ANN, gene expression programming, and multilinear regression approaches for blast-induced ground vibration prediction

초록

Blast-induced ground vibration requires a great deal of consideration in the mining activity because of its effects on the surrounding infrastructure. The impact of blasting activities on natural and man-made structures has received tremendous interest due to growing danger posed by the increase in population leading to more infrastructures close to the quarries. To minimize the combined effects of controllable and uncontrollable blasting parameters on blast-induced ground vibration requires better understanding. In this research, the Buckingham π (pi) theorem (BK), white-box artificial neural network (ANN), gene expression programming (GEP), and multilinear regression (MLR) models are proposed to generate reliable prediction models for the blast-induced ground vibration generated in limestone quarries. One hundred and ninety-one (191) datasets including controllable and uncontrollable blasting parameters were used in developing the models. The datasets were split into three using the ratio 0.70:0.15:0.15 for training, testing, and validation, and the performance of the proposed models was compared. The white-box ANN models performed better than the other models with 0.991, 0.948, and 0.995 coefficients of determinations (R2) for respective training, testing, and validation. A detailed sensitivity analysis conducted using the partial derivative (PaD) method shows that the charge weights per delay, the burden-to-diameter ratio, and distance have more effect on the predicted peak particle velocity (PPV). © 2021, Saudi Society for Geosciences.

키워드