[Rapid Coal Classification Based on Confidence Machine and Near Infrared Spectroscopy]

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Near-infrared reflectance spectroscopy (NIRS) is a widely adopted, safe, and convenient analytical technology.
• Traditional chemical analysis for coal classification is laborious and time-consuming.
• There is a need for rapid and reliable methods for coal characterization in industrial settings.

Purpose of the Study:

• To develop a rapid coal classification method using NIRS.
• To introduce and evaluate the efficacy of a Confidence Machine-SVM (CM-SVM) model for NIRS-based coal analysis.
• To assess the risk and credibility of NIRS analyses through probabilistic methods.

Main Methods:

• Collected near-infrared reflectance spectra from 199 coal samples of four types from Chinese mines.
• Developed machine learning classifiers based on spectral data.
• Implemented a Confidence Machine-SVM (CM-SVM) model for probabilistic classification and risk evaluation.
• Evaluated the CM-SVM model's performance on coal sample classification.

Main Results:

• The CM-SVM model achieved a 95.45% correct grouping rate for coal samples.
• CM-SVM provided confidence and credibility estimations for each prediction.
• The model demonstrated the ability to perform region prediction with predefined error rates based on confidence levels.
• On-line learning capability of CM-SVM showed increasing prediction confidence with more samples.

Conclusions:

• CM-SVM offers a more effective probabilistic approach to NIRS-based coal classification compared to traditional SVM.
• The confidence estimation feature is crucial for quality control in industrial NIRS applications.
• The on-line learning nature of CM-SVM makes it highly suitable for industrial on-line analysis and model improvement.