Developing accurate data-driven soft-sensors through integrating dynamic kernel slow feature analysis with neural networks

A data-driven soft-sensor modelling approach based on dynamic kernel slow feature analysis (KSFA) is proposed in this paper. Slow a extraction method that aims to extract slowly varying features can capture the driving forces behind data. However, there are situations where linear SFA (LSFA) cannot due nonlinear relationships between and input signals. KSFA extension of LSFA utilises trick map inputs into higher-dimensional space. Extracting improve performance by utilising as neural network, which provides information key underlying trends, with added benefit noise reduction. Combining network further improves for cases outputs present. The effectiveness first demonstrated numerical example, theoretical advantages be easily observed. It then applied benchmark simulated industrial fed-batch penicillin process.