The rapid and comprehensive determination of coal quality is of great significance for the optimization of boiler combustion and the digital transformation of coal-fired power plants. Laser-induced breakdown spectroscopy(LIBS) has the potential to be applied effectively in the rapid determination of coal quality. In order to meet the application goal of rapid coal inspection

46 sets of spectral data of coal samples from different power plants were collected by the experimental device of coal particle flow LIBS

and the research of simultaneous rapid inspection of multiple indicators of coal quality by combining LIBS with machine learning was carried out systematically. In view of the considerable spectral fluctuations observed in the particle flow state

the number of single-pulse acquisitions was optimized. In addition

invalid spectral screening

spectral averaging and spectral normalization data preprocessing methods were established. Furthermore

four machine learning algorithms(PLSR

SVR

PSO-SVR

and LSTM) and four spectral feature inputs(full spectra

eigenbands

intensity integration

and PCA extraction) were compared in terms of their performance in predicting multiple indicators of coal quality. The results demonstrate that the uncertainty of the spectral signals can be maintained at a maximum of 5% when 200 single-pulse spectra are collected for spectral averaging in a single test. The PSO-SVR algorithm exhibits the most optimal prediction performance in the quantitative analysis of coal quality indicators

and the PCA algorithm reduces the dimensionality of the spectral data

which reduces the amount of model computation and at the same time improves the prediction performance of the model

and the model established by combining both of them has the best performance

the root mean square error(RMSEP) of the coal heat content is 0.289 MJ/kg

and the mean absolute error(MAE) is 0.231 MJ/kg. The coal carbon mass fraction

ash content and volatile matter content are also predicted satisfactorily

with the RMSEP of 0.987%

1.310% and 1.612%

and the MAE of 0.839%

1.014%

and 1.033%

respectively. The results show that

combined with appropriate machine learning algorithms

the LIBS technique can achieve simultaneous accurate and rapid determination of multiple indicators of coal quality

which has a broad application prospect in the scenario of efficient and clean coal utilization.