Partitioned LSSVR modeling method for self adaption under multi operating conditions

Gangquan Si, Zhang Guo, Jianquan Shi, Hong Gao

Abstract: The real industrial machines always work under several certain operating conditions, which leads to an observation with heteroscedasticity. The estimation solution for unmeasurable parameters is based on LSSVR which focus on the total range so that they can't perform well in some operating conditions. In this paper, a partitioned LSSVR modeling method is proposed to improve the performance under multi operating conditions. The method uses the partition of datasets and the combination of partitioned parts so that the different variance levels on each condition can be concerned. Simulation results show that the proposed method is able to self-adapt under different operating conditions thus each condition can get a better performance, especially the condition with a lower variance level.