A Soft Sensor Method Based on Just in Time Learning and Multi-Model Ensemble GPR

For some difficulties to realize real-time measurements of vital variables, a novel soft sensor based on the just in time learning (JITL) and ensemble learning Gaussian process regression (GPR) modeling was proposed. Firstly, different sub-blocks were created by principal component decomposition. Then a set of JITL-GPR models were developed based on various sub-blocks. Finally, some better results from JITL-GPR models were combined to adaptively obtain the prediction result. The proposed soft sensor was applied to the industrial rubber producing process. Result verifies the superiority of the proposed method.