An optimal sparseness approach for least square support vector machine

Abstract

Least square support vector machine (LSSVM) is a well accepted process modeling technique. However, it has an instinct shortcoming as that the solution is lack of spareness. In this paper, a particle swarm optimization (PSO) based optimal spareness approach for LSSVM is proposed and validated. The spareness of LSSVM is firstly formulated as an optimization problem, where pruning percentage of the training data set is taken as the optimization variable. And then, PSO is employed to solve the spareness problem. A LSSVM model for carbon content in fly ash of utility boiler is used for algorithm validation. Long term operation data of a 600MW boiler is collected for comparison studies. The presented results convince that the PSO based optimal spareness approach exceeds the classical method, and is capable of converging to an optimal support vector set.