Abstract: Currently, most of the coal calorific value prediction models are based on the traditional linear regression model.Also, most of them are based on proximate analysis such as ash, volatile and fixed carbon, while less consideration is given to the effect of elemental analysis on calorific value prediction.Therefore, these methods have low accuracy and narrow application range for predicting coal calorific value.To overcome the shortcomings of traditional linear regression model in coal calorific value prediction, a nonlinear modeling and prediction method by support vector regression (SVR) coupled with mean impact value (MIV) was proposed.By using 60 sets of coal quality analysis sample data from China and considering the influence of proximate analysis and elemental analysis on calorific value, the characteristic variables of influencing the calorific value were selected.Based on this, the nonlinear predication model of coal calorific value was built.The results show that for 60 sets of coal quality analysis sample data from China, only ash and carbon content show a certain linear correlation with calorific value in the indexes of proximate analysis and elemental analysis, the other indexes show a poor linear correlation with calorific value.Thus, nonlinear modeling method should be given priority in coal calorific prediction.After the selection of the characteristic variables of influencing the calorific value, it is found that the influence of moisture, ash and volatile on calorific value is greater in the proximate analysis of coal, while the effect of only carbon content on calorific value is more significant in elemental analysis.When predicting calorific value with above four significant indexes as characteristic variables, the prediction accuracy of SVR method is improved significantly, and the correlation coefficient is 98.27%, which is much higher than other linear regression models.In addition, the prediction results of calorific value based on linear and nonlinear models show that the closer the coal sample source area is, the smaller the difference of prediction accuracy of each method is.It is appropriate to select different predict methods according to the regional differences of samples.