Abstract: Coal ash content is one of the key indicators for evaluating coal quality. The content and properties of ash significantly impact combustion equipment,the environment,and subsequent processing and utilization. To address the issues of lagging and labor-intensive methods in current coal ash detection,an intelligent prediction algorithm for coal ash content based on X-ray Fluorescence (XRF) spectroscopy combined with Preprocessing (PRE) and Partial Least Squares (PLS) regression is proposed. The process involves inputting spectral data of coal samples obtained via XRF technology into the PLS main model for an initial ash content prediction. Relevant correction parameters are then fed into a parameter compensation optimization model. The final predicted ash content value is derived by summing these two results. Experimental results indicate that,compared to partial least squares regression and neural network (NN) regression models,the PRE-PLS model exhibits the smallest fluctuation range in relative error. Compared with partial least squares regression and neural network regression models, the determination coefficient of the PRE-PLS model is 0. 995 1, the root mean square error is 0. 941 1, and the mean absolute error is 0. 733 2%. This indicates that this model has high accuracy and is capable of performing on-site detection tasks, providing reliable guidance for production. 

Key words: X-ray fluorescence spectroscopy (XRF),coal,ash content prediction,partial least squares (PLS), spectral preprocessing