Slagging in the furnace is one of the key factors affecting the reliable operation of thermal power units and gasification process. The safety and economy of the operation of thermal power units and gasifiers can be improved by accurate and timely measurement of ash melting temperature. However, there are many uncertain factors in the process of ash melting temperature measurement. The establishment of ash melting temperature prediction method cannot only verify the reliability of test data, but also replace the complicated test to a certain extent. The composition, classification methods, similarities and differences of coal ash and biomass ash were discussed, and the effects of different oxides on ash fusibility were summarized. Three main methods for predicting the melting temperature of coal and biomass ash, including empirical formulas, machine learning models, and multivariate phase diagrams, were described, and the advantages, disadvantages, and applicability of each method were analyzed. It is considered that the empirical formula is more suitable for the coal ash data set with single variety and small quantity, but it is not suitable for the prediction of biomass ash melting point. The machine learning model has good prediction effect on coal ash and biomass ash, but it is more difficult to model, which requires more training sample data. The prediction of ash melting temperature based on phase diagram is limited by the ash fusibility test method, and the prediction effect is not better than the empirical formula and machine learning model, but it has good prediction accuracy for four typical coal types, and the biomass ash has more special samples than coal ash. Further research is needed to determine whether it can be used for the prediction of biomass ash melting temperature. In the future, it is possible to consider building K nearest neighbor regression, random forest and other more outstanding models to solve regression problems and expand biomass database samples to improve the accuracy and generalization ability of the prediction model.