Coking coal in coal chem. enterprises presents a challenge due to its diverse types, wide sources, and variable quality, influenced by varying degrees of metamorphism and physicochem. properties.These differences not only impact coal quality but also hinder accurate anal.Rapid and precise coal quality detection amidst diverse coal types is crucial for maintaining stable production and ensuring coke quality.This study employs near-IR spectroscopy (NIRS) and X-ray fluorescence spectroscopy (XRF) fusion spectroscopy anal., along with Principal Component Anal. (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE), for data dimensionality reduction and visualization to devise a coal sample classification strategy.Subsequently, Support Vector Machine (SVM) is employed for automatic coal sample classification based on this strategy.Finally, Partial Least Squares Regression (PLSR) is used to establish regression models and evaluate their performance in predicting coal quality.Results show that classified regression model achieves R2 values of 0.9987, 0.9955, and 0.9997 for ash content, volatile matter, and sulfur content, with corresponding root mean square error for prediction (RMSEP) of 0.31 %, 1.34 %, and 0.05 %, and the mean absolute relative error for prediction (MARDP) of 2.48 %, 3.58 %, and 3.57 %, resp.Compared to the unclassified model, there is a significant enhancement in prediction accuracy.The classification and modeling method proposed herein effectively improve the accuracy of coal quality anal. in complex coal type scenarios, crucial for industries like coal chem. engineering to enhance production efficiency and optimize coal resource utilization.