Soft sensor modeling for mill output of direct fired system based on improved least squares support vector machines

According to the problem that the direct-fired system with duplex inlet and outlet ball pulverizer is a large time delay and strongly nonlinear system,its mill output is difficult to be directly measured.In this paper,a better improvement algorithm was proposed based on the Suykens' sparseness method of least squares support vector machines(LS-SVM).This new method is that when deleting some too large or too small training samples,the large variety-rate data were also deleted.This has simplified the LS-SVM model and avoided the bad sample's effect.The improved LS-SVM was used to establish the soft sensor model of mill output in direct-fired system with duplex inlet and outlet ball pulverizer.The before and after improvement LS-SVM algorithm was respectively simulated to verify it's precision.The mean square error after improvement is 0.0227,reduced 0.0119 than before,and it's learning speed is faster.So it is more suitable to study on-line.