A least squares support vector machine(LS-SVM)and grey model(GM)based dynamic soft sensor method for ball mills was proposed.By analyzing the factors affecting the coal level

the auxiliary variables of the soft sensor model were determined.The LS-SVM based soft sensor static model was established

of which the results were compared with that of the actual values.Thus the time measurement errors sequence was obtained and then modeled and predicted by the GM.Finally

the predictive error results were combined with the static model to realize dynamic correction.Application example shows this method can reflect the trend and dynamic characteristics of coal level effectively

which has a higher accuracy and applicability than the single LS-SVM model.