The Analysis of Coal Calorific Value Based on Neural Networks and Hyperspectral Images with Limited Samples

The calorific value of coal is a critical parameter for evaluating its quality and economic utility. Traditional methods for calorific value determination primarily rely on chemical analysis, which are time-consuming and costly. Given the advantages of hyperspectral imaging (HSI) in rapid and non-destructive material analysis, this paper proposes a novel approach for coal calorific value prediction by integrating HSI with neural networks. Spectral data of coal samples were first acquired using a hyperspectral imaging system and subjected to preprocessing. To enhance prediction performance, the neural network model was optimized using a combination of genetic algorithm (GA), momentum-based gradient descent, and adaptive learning rate strategies. Experimental results demonstrate that the optimized model achieved a root mean square error (RMSE) of 0.7102 MJ/kg, a mean absolute error (MAE) of 0.5940 MJ/kg, and a coefficient of determination (R²) of 0.9871. These findings confirm the feasibility and effectiveness of the proposed method for rapid and non-destructive coal calorific value analysis, offering a promising alternative for coal quality assessment.