A coal ash fusion temperature model is constructed based on support vector machine(SVM). The compositions of coal ash are employed as the inputs and the ash fusion temperature is the output. A series of improvement is made on basic ant colony optimization(ACO) and it is used to optimize the parameters of the SVM model. The coal ash fusion temperature is predicted by the ACO-optimized SVM model. Some experiments are performed to compare the predicted and the measured temperature and the results show the ACO-optimized SVM model can achieve better predicting performance. The advantages of SVM model, such as small sampling, fast computing speed and real-time processing and predicting are also displayed.