The ash fusion temperature (AFT) of coal ash is a key factor that influences the slagging process during coal gasification. However

due to the complex influencing factors and the coal characteristics differences from different mines

predicting AFT remains a challenge. In this paper

2338 sets of production data from various mines in China were preprocessed

and typical machine learning and ensemble learning methodologies coordinated with Bayesian optimization were established to obtain an accurate and robust model. The results demonstrate that the ensemble learning models namely extreme gradient boosting and gradient boosting decision tree

exhibited the lowest root mean squared error of approximately 13.00

mean absolute error at 6.93 and 7.11

respectively

and determination coefficient R2 at 0.90. The Shapley additive explanation interpretability analysis was implemented to reveal the contribution of each feature to the AFT. This work is significant for accurately predicting the AFT of coal ash. By providing feature importance

it helps explain the predicted output and offers valuable suggestions and references for the coal blending scheme.