Soft sensor modeling of mill output in direct fired system based on improved FIR filter and least squares support vector machines

Abstract EN

According to the multivariable, strongly nonlinear and large time delay system, the computing time of least squares support vector machines (LS-SVM) as a soft sensor modeling method is longer because of its lacking sparseness.

It goes against on-line realization.

In this paper, FIR digital filter was used because of its linear phase properties.

Firstly, finite impulse response FIR) digital filter was optimized by Remez) algorithm.

Simulation results show that the optimization effect is better.

Secondly, the improved algorithm was combined with LS-SVM and formed a new soft sensor modeling method, FIR-LSSVM.

Lastly, FIR-LSSVM was used to establish the soft sensor model of mill output of direct-fired system with duplex inlet and outlet ball pulverizer in power plant.

Field tests show that the learning speed of FIR-LSSVM is faster and the error is less.

FIR-LSSVM is more suitable online learning.