Abstract:Peak particle velocity (PPV) is an important criterion for assessing the risk level of ground vibration induced by mine blasting. Based on this criterion, many efforts to predict accurately and mitigate PPV have been studied. In this paper, a novel integration approach based on the extreme learning machine (ELM) and multi-verse optimization (MVO), named the MVO–ELM model, was proposed to predict PPV. MVO was applied to initialize the number of populations and optimize the weights of the ELM model. For this, 137 blasting events in a copper mine in China with blasting parameters and rock properties were collected to test the proposed model. They were then divided into training and testing subsets for developing and testing the model, respectively. Different training/testing ratios of data subsets (e.g., 80/20 and 70/30), different topology networks (e.g., 22 neurons and 25 neurons), and various activation functions (e.g., elu, relu, tanh, and sigmoid) were considered to determine the best data subset ratio and topology network of the proposed model. To evaluate the enhanced accuracy of the proposed MVO–ELM, the traditional ELM, multi-perceptron (MLP) neural network, and the USBM (United States Bureau of Mines) models were applied to predict PPV for comparison with the optimized MVO–ELM model. The results revealed that the MVO–ELM model was the best one, with data subset ratio of 80/20, topology network with 25 hidden neurons, and "elu" activation function. The performance metrics also indicated that the optimized MVO–ELM model provided high-caliber performance compared with the ELM, MLP, and USBM models, i.e., RMSE (root-mean-squared error) = 0.196 and 0.374, R2 (determination coefficient) = 0.980 and 0.943, MAPE (mean absolute percentage error) = 0.206 and 0.670, VAF (variance accounted for) = 98.048 and 94.350, on the training and testing data subsets, respectively. Moreover, the accuracy of the optimized MVO–ELM model was improved by approximately 36% compared to the traditional ELM model. In contrast, the ELM, MLP, and USBM yielded lower performances, i.e., RMSE = 0.846, 0.657, 1.008, R2 = 0.638, 0.782, 0.558, MAPE = 1.242, 0.653, 0.691, and VAF = 63.787, 78.126, 53.406, on the training dataset; RMSE = 1.020, 0.718, 0.834, R2 = 0.619, 0.827, 0.815, MAPE = 1.219, 1.090, 0.868, and VAF = 58.284, 79.191, 75.565, on the testing data subset. The results also revealed that the maximum explosive charged per delay, monitoring horizontal distance, and rock mass integrity coefficient significantly affected PPV predictions.

State-of-the-art review of machine learning and optimization algorithms applications in environmental effects of blasting

Knowledge mapping of research progress in blast-induced ground vibration from 1990 to 2022 using CiteSpace-based scientometric analysis

Optimization of Blasting Parameters Considering Both Vibration Reduction and Profile Control: A Case Study in a Mountain Hard Rock Tunnel

Machine-Learning-Aided Determination of Post-blast Ore Boundary for Controlling Ore Loss and Dilution

Application of Artificial Neural Network (ANN) for Prediction and Optimization of Blast-Induced Impacts

A rapid evaluation method of blasting effect based on optimized image segmentation algorithm and application in engineering

Estimating Flyrock Distance Induced Due to Mine Blasting by Extreme Learning Machine Coupled with an Equilibrium Optimizer

Novel Extreme Learning Machine-Multi-Verse Optimization Model for Predicting Peak Particle Velocity Induced by Mine Blasting

Towards lightweight excavation: Machine learning exploration of rock size distribution prediction after tunnel blasting

Predicting ground vibration during rock blasting using relevance vector machine improved with dual kernels and metaheuristic algorithms

Rockburst prediction and prevention in underground space excavation

Enhancing rock fragmentation assessment in mine blasting through machine learning algorithms: a practical approach

Hybrid Metaheuristic Optimization Algorithms with Least-Squares Support Vector Machine and Boosted Regression Tree Models for Prediction of Air-Blast Due to Mine Blasting

Emerging trends in drilling and blasting technology: concerns and commitments

Blastability evaluation for rock mass: review and new tendency

Performance evaluation of hybrid WOA-XGBoost, GWO-XGBoost and BO-XGBoost models to predict blast-induced ground vibration

Developing a hybrid model of Jaya algorithm-based extreme gradient boosting machine to estimate blast-induced ground vibrations

Novel Soft Computing Model for Predicting Blast-Induced Ground Vibration in Open-Pit Mines Based on Particle Swarm Optimization and XGBoost

Numerical analysis on dynamic response and damage threshold characterization of deep rock mass under blasting excavation

Applications of Two Neuro-Based Metaheuristic Techniques in Evaluating Ground Vibration Resulting from Tunnel Blasting

Application of machine learning, deep learning and optimization algorithms in geoengineering and geoscience: comprehensive review and future challenge