Predicting Blast-Induced Ground Vibration in Open-Pit Mines Using Vibration Sensors and Support Vector...

Area of Science:

• Geotechnical Engineering
• Computational Intelligence
• Vibration Analysis

Background:

• Blast-induced ground vibration, quantified as peak particle velocity (PPV), is a critical factor in mining and construction.
• Accurate PPV prediction is essential for environmental impact assessment and structural safety.
• Existing prediction models often require refinement to account for complex geological and blasting parameters.

Purpose of the Study:

• To develop and evaluate novel hybrid models for predicting PPV using evolutionary algorithms and Support Vector Regression (SVR).
• To compare the performance of four evolutionary algorithms: Particle Swarm Optimization (PSO), Genetic Algorithm (GA), Imperialist Competitive Algorithm (ICA), and Artificial Bee Colony (ABC).
• To investigate the efficacy of three kernel functions (Linear, RBF, Polynomial) within the SVR framework for PPV prediction.

Main Methods:

• Utilized vibration sensors to collect 125 blast-induced ground vibration datasets from a limestone quarry in Vietnam.
• Developed 12 hybrid models by integrating GA, PSO, ICA, and ABC with SVR, each using Linear, RBF, or Polynomial kernel functions.
• Employed statistical metrics (R², RMSE, MAE) and ranking methods for comprehensive model evaluation.

Main Results:

• The Genetic Algorithm (GA) demonstrated superior performance as an evolutionary algorithm when integrated with SVR.
• The Radial Basis Function (RBF) kernel was identified as the most effective kernel function for the GA-SVR model.
• The GA-SVR-RBF hybrid model achieved the highest accuracy in predicting PPV among all developed models.

Conclusions:

• The GA-SVR-RBF model is proposed as a highly effective and reliable technique for estimating blast-induced ground vibration (PPV).
• Hybrid models combining evolutionary computation and machine learning offer significant potential for geotechnical engineering applications.
• The study highlights the importance of algorithm and kernel function selection in optimizing predictive model performance.