Improvement and AppLication of MKSVM

Grinding production rate (GPR) is a vital index of grinding process. Getting accurate and timely information of GPR is the premise of enhancing grinding efficiency and conducting optimization control. However, for complexity of grinding process, there is no effective method to on-Line predict GPR. On the basis of soft sensor principle, a new sCheme that applying improved mixed-kernel support vector machine (MKSVM) to predict GPR is presented. Meanwhile nesting termination condition genetic algorithm (NTCGA) is proposed to improve the deficiency of MKSVM that there is no effectual way to select parameters of MKSVM´s kernel. Furthermore comparison experiments of prediction between the improved MKSVM and traditional support vector machines (SVMs) and radial basis function (RBF) network are conducted. Simulation results show the prediction accuracy of GPR by using the improved MKSVM can meet the demand of practical appLications. And performance of improved MKSVM is much better than other SVMs and RBF network.