Development Of An Adaptive Soft Sensor Based On FCMILSSVR

Abstract- Facing with dynamic environment of industrial plants involves us design soft sensors capable of online learning. To response this requirement an adaptive soft sensor based on a combination of Least Square Support Vector Regression LSSVR with Fuzzy C-Means FCM clustering is proposed in this paper. In this approach first the samples are divided into several partitions. Consequently for each partition we develop a local model using a new formulation of LSSVR which enables incremental learning. The proposed method is implemented on a chemical plant and compared with the online Support Vector Regression SVR algorithm. Simulation results indicate that the proposed method improves the generalization ability of soft sensor and the computation time decreases to a large extent in comparison to the online SVR.