Research On Non-parametric And Semi-parametric Soft Sensor Modeling Method In Fermentation Process

Fermentation process is a complicated and multi-parametric strong non-linear process, each parameter can directly impact the generation of fermentation product. Modeling on soft sensor of fermentation process play a key role on fermentation process detection and product concentration analyzation. The estimation precision of the soft sensing method in the present fermentation process is influenced by the cognitive level of the fermentation process and the structure of the soft sensor model. Non-parametric regression can leverage untreated historical data directly, estimation under un-limited modeling structure and non-requirement on sample distribution shows excellent imitative effect on non-linear system. Semi-parametric regression method bases on non-parametric regression method and process mechanism also makes modeling more accurate. Therefore, research on the soft sensor modeling method based on non-parametric and semi-parametric regression fermentation process is great significant and value on promoting the online estimation of biological parameters precision of fermentation process soft sensor.This paper research on non-parametric regression method, present the fermentation process soft sensor method based on non-parametric regression which based on the fermentation process characteristic of much input and output also avoid artificially simplifying calculation steps to promote calculation effectiveness; research on semi-parametric regression method with fermentation process which runs system identification with fermentation process mechanism also modeling by non-parametric regression on the parameters residual to get the modeling of semi-parametric regression soft sensor in fermentation process; improve the method of semi-parametric regression which takes the fermentation process batch as the samples to build the model that more simple than traditional semi-parametric regression method and makes the modeling more accurate.As an experimental result, the method shows good accuracy. Soft sensor modeling of fermentation process based on semi-parametric regression can estimate the product concentration. It also shows faster rate of convergence and better reliability.