Multi-Energy Spectroscopic Fusion and Targeted Feature Engineering for Accurate Coal Carbon Quantification.

Area of Science:

• Analytical Chemistry
• Environmental Science
• Spectroscopy

Background:

• Coal combustion is a significant source of carbon dioxide (CO2) emissions, necessitating precise carbon quantification for effective emission assessment and mitigation strategies.
• Accurate analysis of carbon content in coal is crucial for environmental monitoring and quality control.

Purpose of the Study:

• To develop and evaluate a trimodal fusion prediction system for precise carbon estimation in coal.
• To integrate multi-energy laser-induced breakdown spectroscopy (LIBS) and laser-induced plasma acoustic (LIPA) signals with machine learning algorithms.

Main Methods:

• Quantitative models were established using five standard coal samples with external and internal standard methods.
• Low-level data fusion of LIBS and LIPA signals was evaluated using various machine learning algorithms, with Random Forest showing optimal performance.
• Novel feature extraction methods, including targeted area-preserving PCA and hybrid time-frequency alignment PCA, were developed and applied.

Main Results:

• The baseline model using LIBS spectral features lacked chemical interpretability, excluding carbon-related features.
• Trimodal data fusion incorporating LIBS, LIPA, and multi-energy information significantly enhanced model accuracy, reliability, and generalization.
• The developed system demonstrated improved predictive capability through feature importance ranking and advanced feature extraction techniques.

Conclusions:

• The proposed trimodal fusion system offers a promising approach for accurate CO2 emission monitoring and coal quality assessment.
• Integration of multi-energy LIBS, LIPA signals, and advanced machine learning techniques provides superior carbon analysis compared to traditional methods.