Abstract:

It is difficult setting the kernel parameter which is applied to online identify a time-varing nonlinear process by using least square support vector machine(LSSVM), and such setted one can not automatically adjust to adapt it to the varying process. In view of this situation, an online process identification approach based on time-division regulated kernel parameter LSSVM is proposed. Three LSSVMs are utilized and the whole modeling predicting times are divided into a starting stage and several working periods.  The LSSVM with a smallest sum of predicting errors is selected as a working LSSVM for successive working periods at the end of both the starting stage and each working period. Moreover, kernel parameters are reset for  other two LSSVMs according to heuristic rules, and they are used as comparative LSSVMs for the following working periods. The method is easy to set kernel parameters and has adjustability to a certain extent. The numerical simulation shows the adaptability of the method is better than that of traditional methods statistically.