Accurate Coal Classification Using PAIPSO-ELM with Near-Infrared Reflectance Spectroscopy.

Area of Science:

• Geoscience and Materials Science
• Computational Intelligence and Machine Learning

Background:

• Traditional coal classification methods face limitations in accuracy and efficiency.
• Inefficient coal utilization stems from classification inaccuracies.
• Near-infrared reflectance spectroscopy (NIRS) offers potential for coal analysis.

Purpose of the Study:

• To develop a novel, accurate, and efficient coal classification method.
• To overcome the limitations of traditional coal classification techniques.
• To enhance the utilization of China's vast coal reserves.

Main Methods:

• Collected near-infrared reflectance spectroscopy (NIRS) data from coal samples.
• Applied Extreme Learning Machine (ELM) for initial classification.
• Utilized Particle Swarm Optimization (PSO) to optimize ELM parameters.
• Developed an improved Position-Adaptive Inertia PSO-ELM (PAIPSO-ELM) model.

Main Results:

• The ELM model showed good initial classification performance.
• The PSO-ELM model improved classification accuracy by 9.68% over the basic ELM.
• The PAIPSO-ELM model achieved an additional 2% accuracy improvement without increasing training time.
• The PAIPSO-ELM model demonstrated superior performance in overcoming local optima.

Conclusions:

• The PAIPSO-ELM model provides an effective solution for coal spectral classification.
• This method addresses industrial demands for high accuracy and speed in coal classification.
• The proposed approach enhances the efficient utilization of coal resources.