Enhanced prediction of bolt support drilling pressure using optimized Gaussian process regression

Area of Science:

• Geotechnical Engineering
• Computational Intelligence

Background:

• Accurate prediction of drilling pressure in bolt support systems is crucial for geotechnical engineering.
• Gaussian Process Regression (GPR) shows potential but requires optimization for kernel functions and historical data.
• Variability in GPR prediction outcomes necessitates robust parameter tuning.

Purpose of the Study:

• To develop and validate a novel, optimized GPR method for predicting drilling pressure in underground bolt support systems.
• To enhance the accuracy and efficiency of drilling pressure predictions.
• To identify optimal GPR parameters for practical geotechnical applications.

Main Methods:

• Explored 160 GPR schemes combining 10 kernel functions and 16 historical points.
• Applied hybrid optimization algorithms: Genetic Algorithm-GPR (GA-GPR), Particle Swarm Optimization-GPR (PSO-GPR), and Ant Colony Algorithm-GPR (ACA-GPR).
• Iteratively optimized GPR parameters using the hybrid algorithms.

Main Results:

• PSO-GPR demonstrated superior performance, achieving 80% prediction accuracy with a 1-2 MPa deviation.
• The optimal GPR configuration identified was the RQ kernel function with 18 historical points.
• The optimal combination yielded a Root Mean Square Error (RMSE) of 0.0047246, significantly lower than the least effective combination (RMSE of 0.035704).

Conclusions:

• A robust and efficient prediction system for underground bolt support drilling pressure was developed using PSO-GPR.
• The optimized GPR model significantly improves the accuracy and efficiency of geotechnical support systems.
• The study validates the practical applicability of the PSO-GPR model in real-world drilling operations.