Coal Calorific Value Prediction via Multi-View Transformer

Accurate measurement of coal calorific value is critical for efficient power generation. To address the limitations of conventional models on large-scale, heterogeneous datasets, this study proposes a novel deep learning framework, the Multi-View Transformer (MVFormer), utilizing fused Near-Infrared Spectroscopy (NIRS) and X-ray Fluorescence (XRF) data. The architecture employs a dual-pathway Transformer with a Masked Autoencoder pre-training strategy to enhance feature representation from over 20,000 coal samples. Furthermore, a multi-view fusion mechanism integrates diverse pre-processing perspectives to enhance generalization. Experimental results demonstrate that this approach significantly outperforms traditional Partial Least Squares (PLS) regression and Multilayer Perceptron (MLP) models. These findings validate the framework as a robust and precise solution for real-time industrial coal quality analysis, successfully achieving precise prediction of calorific value on large-scale coal datasets.