Abstract: The gross calorific value (GCV) of coal is an important indicator for coal quality evaluation, which directly affects combustion efficiency and production costs. However, GCV is affected by multiple factors such as element composition and industrial analysis indicators, etc., and there are complex nonlinear relationships between variables, making it difficult for traditional linear regression, empirical formulas, and some machine learning models to accurately model. Although deep learning methods (such as LSTM and Transformer) can capture complex modes, their "black box" characteristics limit industrial applications. In order to accurately predict the gross calorific value (GCV) of coal, a prediction model based on Pearson correlation coefficient and Kolmogorov-Arnold network is proposed. Through feature selection and nonlinear modeling, the prediction accuracy is significantly improved, providing technical support for coal quality evaluation and combustion efficiency optimization. The experimental results show that this model outperforms traditional methods in terms of MAE, MSE, and other indicators.