Abstract: An online forecasting model based on self-tuning support vectors regression (SVR) for zinc output in imperial smelting furnace is put forward with the aim of maximizing zinc output by adjusting operational parameters. In this model, the mathematical model of SVR is converted into the same format as that of support vector machines for classification. A simplified sequential minimal optimization (SMO) for classification is applied to train the regression coefficient vector a-a* and threshold b. Penalty parameter C can be tuned dynamically with the forecasting result until the training process ends. The online forecasting algorithm for zinc output is also presented. In spite of a relatively small industrial data set, the simulation result shows that the effective error is less than 10% with a remarkable performance of real time.