Abstract: Laser-induced breakdown spectroscopy (LIBS) technology faces challenges in coal quality analysis due to multi-physical field coupling interferences, significant matrix effects, and limited measurement accuracy and reproducibility. To address these issues, a method integrating RSM system optimization with ANN-SHAP interpretable analysis is proposed for online prediction of coal industrial analysis indicators using LIBS. First, focusing on the nonlinear characteristics of LIBS analysis, a multi-objective fusion response function considering both spectral signal quality and stability was constructed. A multi-factor, multi-level Box-Behnken design (BBD) was employed, and response surface methodology (RSM) was applied to establish mathematical regression models linking key LIBS system parameters—including laser energy, laser frequency, and spectrometer delay time—to the response function, thus completing system optimization and determining the optimal experimental conditions for acquiring high-quality spectral data.Second, sample data comprising 120 sets of four-channel spectra over a wavelength range of 160–960 nm were collected, and preprocessing steps including baseline correction, denoising, feature selection, and internal standardization were applied. Subsequently, spectral data modeling was performed using a genetic algorithm (GA) combined with an artificial neural network (ANN), and the model performance was evaluated. Experimental results demonstrated that the calibration models for calorific value, ash content, moisture, and volatile matter performed well on the test set, with coefficients of determination R2 of 0.971, 0.987, 0.975, and 0.968, root mean square errors RMSE of 0.58 MJ/kg, 1.23%, 0.78%, and 0.98%, and prediction repeatability r of 0.10%, 0.21%, 0.19%, and 0.24%, all meeting national standard requirements.Finally, to address the “black-box” nature of the artificial neural network, SHAP analysis was introduced to provide a global interpretable analysis of the model, revealing the decision mechanisms and the nonlinear contributions of spectral features to the coal quality indicators, thereby enhancing the interpretability and engineering reliability of the model.