Hybrid machine learning with metaheuristic optimization for predicting blast-induced peak particle velocity in openpit mines

Blast-induced ground vibration remains one of the most critical environmental and safety challenges in openpit mining. Excessive vibration can damage nearby structures, destabilize slopes and create operational and community concerns. This study presents a hybrid machine learning (ML) framework that integrates random forest (RF) with three advanced metaheuristic optimization algorithms — whale optimization algorithm (WOA), particle swarm optimization (PSO) and grey wolf optimizer (GWO) — for accurate prediction of peak particle velocity (PPV). Using 175 blasting events collected from the Jayant opencast coal mine in India, the proposed WOA-RF model achieved superior predictive performance compared with conventional empirical equations and baseline ML approaches. The developed framework also includes a practical graphical user interface (GUI) that allows engineers to perform real-time PPV prediction and blast design optimization. The proposed methodology can help mine operators improve blasting efficiency, minimize environmental impacts and enhance operational safety in real-world mining projects. Background Drilling and blasting are essential operations in large-scale openpit mining. However, blasting generates undesirable environmental effects such as flyrock, air overpressure, dust and ground vibration (Fig. 1). Among these, ground vibration is considered the most critical because it can propagate over long distances and potentially damage nearby infrastructure. PPV is widely accepted as the standard parameter for evaluating blasting-induced ground vibration due to its strong correlation with structural damage. Traditionally, PPV prediction has relied on empirical scaled-distance equations. Although these equations are simple and widely used, they often fail to capture the complex nonlinear interactions between geological conditions, blast design parameters and explosive characteristics. Recent advances in ML have significantly improved predictive capabilities for blasting applications. ML models can incorporate multiple influencing factors simultaneously and identify nonlinear relationships within complex data sets. However, the predictive performance of ML models strongly depends on proper hyperparameter optimization. Metaheuristic optimization algorithms provide an effective solution for improving ML model performance by systematically searching for optimal parameter combinations. This study therefore developed three hybrid models — WOA-RF, PSO-RF and GWO-RF — and conducted a comprehensive comparative evaluation of their predictive capability, robustness and practical applicability. Fig. 1 Schematic illustration of blasting-induced hazardsand ground vibration wave propagation mechanisms. Data set and methodology The data set used in this study consisted of 175 blasting events collected from an opencast coal mine. Eleven important input variables were selected after correlation analysis and multicollinearity screening: hole diameter (D), hole depth (HD), burden (B), explosive per hole (q), distance from blast face (Dis), Young’s modulus (E), Poisson’s ratio (PR), p-wave velocity (PV), blastability index (BI), velocity of detonation (VOD) and density of explosive (DOE). The data set was divided into training (80 percent) and testing (20 percent) subsets using stratified random sampling to ensure statistical consistency between data sets. Initially, several ML models, including decision tree (DT), support vector regression (SVR), kernel extreme learning machine (KELM), artificial neural network (ANN) and RF, were evaluated. Among these, RF demonstrated the best baseline performance and was therefore selected as the core predictive model. Three metaheuristic optimization algorithms — WOA, PSO and GWO — were subsequently integrated with RF. These algorithms optimized key RF hyperparameters, including maximum tree depth and minimum leaf size using mean absolute error (MAE) as the objective function. Results and discussion The results clearly demonstrate the superiority of the hybrid optimization framework over both conventional ML methods and traditional empirical formulas. Among the developed models, WOA-RF consistently achieved the best predictive performance. The model produced high prediction accuracy for both training and testing data sets, indicating strong generalization capability and reduced overfitting risk. Although PSO-RF achieved slightly higher training performance, its testing accuracy deteriorated considerably, suggesting reduced robustness under unseen conditions. Fig. 2 Regression performance chart of different model trainingand test set fittings. The comparison with traditional empirical equations further highlighted the advantages of the proposed hybrid framework. Empirical models produced relatively low prediction accuracy, with R² values generally below 0.40, whereas the WOA-RF model achieved an R² exceeding 0.92 (Fig. 2). The optimization behavior of the three algorithms also differed substantially. GWO and PSO converged rapidly during early iterations but showed instability under certain population sizes. In contrast, WOA demonstrated more stable and gradual convergence, ultimately providing the best balance between prediction accuracy and computational efficiency. To further evaluate model robustness, Monte Carlo simulations were conducted under varying noise levels. Even at 10 percent noise contamination, the WOA-RF model maintained strong predictive capability with only marginal reductions in R² and variance inflation factor (VAF) values. This demonstrates the model’s reliability under realistic field uncertainties where noisy and imperfect data are common. Sensitivity analysis using SHAP (Shapley additive explanations) revealed that Dis was the most influential parameter affecting PPV prediction. HD, q, B and E also showed significant influence. The SHAP results were consistent with established physical principles governing vibration attenuation and energy propagation in rock masses. Practical application through GUI development To facilitate industrial implementation, a user-friendly GUI was developed, incorporating all three optimized RF models. The interface allows engineers to upload local blasting data sets, train customized predictive models, perform sensitivity analysis, generate real-time PPV predictions and optimize blast design parameters. The GUI was designed specifically for field engineers with limited programming expertise, thereby bridging the gap between advanced ML research and practical mining applications. The interface also integrates SHAP-based interpretability tools, enabling users to understand the influence of different blast parameters on predicted vibration levels. This functionality supports safer and more informed blasting decisions in operational environments. Conclusions This study developed a robust hybrid ML framework for predicting blast-induced PPV in openpit mines. The integration of RF with metaheuristic optimization algorithms significantly improved predictive performance compared with conventional empirical equations and baseline ML models. Among the developed models, WOA-RF achieved the highest prediction accuracy and best generalization capability. The SHAP-based sensitivity analysis further improved model interpretability by identifying the dominant factors influencing vibration propagation. In addition, the developed GUI enhances the practical usability of the proposed framework by enabling real-time prediction and blast design optimization. The proposed methodology offers a practical, accurate and field-deployable solution for vibration management in mining operations and can contribute significantly toward safer and more sustainable blasting practices. Selected references Khandelwal M (2008) Evaluation and prediction of blast induced ground vibration and frequency for surface mine-a neural network approach. PhD thesis, Indian Institute of Technology, Bombay, IndiaKhandelwal M (2010) Evaluation and prediction of blast induced ground vibration using support vector machine. Int J Rock Mech Min Sci 47(3):509–516

Hybrid machine learning with metaheuristic optimization for predicting blast-induced peak particle velocity in openpit mines

Authored by: Jiehao Gu, Yulin Zhang, Chuanqi Li, Jian Zhou, Moshood Onifade and Manoj Khandelwal

To view this article you must be a member of SME/UCA.

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

To view this article you must be a member of SME/UCA.

Chinese brokerages push for LME membership to expand global metals role

Integrating carbon capture into copper comminution: A combined techno-economic assessment and carbon footprint analysis

Related Articles

Rare earth elements in the global economy: Usage, recovery and the quest for supply security — A review

A systematic evaluation of localization signals in the advancing openpit mine Aitik

From foundation to future: Revisiting AI integration in mining engineering education through current perspectives of students, educators and industry

Iron ore tailings reprocessing by electrostatic separation plus two-stage froth flotation

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Mining Engineering’s Weekly Rundown: Episode 73

Get your SME Foundation VIP Access Conference Pass to MINEXCHANGE 2027

“Check the box” and help support the SME Foundation

Use Single Sign On