A Novel Hybrid Model for Predicting Blast-Induced Ground Vibration Based on k-Nearest Neighbors and Particle Swarm...

Area of Science:

• Engineering
• Artificial Intelligence
• Geotechnical Engineering

Background:

• Blast-induced ground vibration, specifically peak particle velocity (PPV), is a critical factor in surface mining operations.
• Accurate estimation of PPV is essential for mitigating potential damage and ensuring safety.
• Existing methods for PPV estimation include empirical techniques and machine learning models, each with limitations.

Purpose of the Study:

• To develop and propose a novel artificial intelligence technique, PSO-KNN, for enhanced estimation of blast-induced ground vibration (PPV).
• To optimize the hyper-parameters of the k-nearest neighbors (KNN) algorithm using particle swarm optimization (PSO).
• To evaluate the performance of the proposed PSO-KNN models against established benchmarks.

Main Methods:

• Development of three PSO-KNN models (PSO-KNN-Q, PSO-KNN-T, PSO-KNN-C) by optimizing KNN kernel functions (Quartic, Tri weight, Cosine) with PSO.
• Utilizing maximum explosive per blast delay (W) and PPV measurement distance (R) as input parameters.
• Comparing the performance of PSO-KNN models against Random Forest (RF), Support Vector Regression (SVR), and an empirical technique using 152 blasting events.
• Evaluating accuracy using Root Mean Square Error (RMSE), R-squared (R²), and Mean Absolute Error (MAE).

Main Results:

• The particle swarm optimization (PSO) algorithm significantly enhanced the efficiency of the PSO-KNN models.
• The PSO-KNN-T model achieved the best performance among the proposed models and benchmarks.
• Specific performance metrics for PSO-KNN-T: RMSE = 0.797, R² = 0.977, MAE = 0.385.
• All PSO-KNN models outperformed the benchmark models (RF, SVR, empirical).

Conclusions:

• The PSO-KNN technique, particularly the PSO-KNN-T model, is a highly effective tool for estimating blast-induced ground vibration (PPV).
• This approach offers a robust and accurate method for predicting PPV in practical blasting scenarios.
• The developed models can aid in reducing unwanted environmental impacts associated with PPV in surface mines.