A Novel Adaptive Soft Sensor Using Multiple Heterogeneous Model Ensemble Learning

In chemical processes, systems are rarely pure linear or nonlinear, but rather present mixed-linear-nonlinear characteristic generally. Therefore, a mixed-linear-nonlinear adaptive soft sensor using multiple heterogeneous models ensemble learning is proposed allowing more effective way to describe mixed-linear-nonlinear characteristic, while the combining models can demonstrate different behavior because of the diversity of different models. Furthermore, due to its inherent redundant design, a soft senor can be made robust. Additionally, RPLS (Recursive partial least squares) is implemented for the weighted combination method to deal with time-varying behavior of processes. The usefulness of the proposed method was demonstrated through a case study of a wastewater treatment process.