Soft sensor modeling based on GRNN for biological parameters of marine protease fermentation process

A soft sensor model based on generalized regression neural network (GRNN) is presented to be aimed at solving the problem that the key biological parameters in the microbial fermentation process are difficult to be real-time online measurement. GRNN contains a special linear output layer, which makes that its network structure possesses the adaptive certainty and output has noting to do with the initial weights, etc. The marine protease fermentation process is taken as an example, firstly through the analysis of the mechanism of marine protease fermentation process, the auxiliary and leading variables of soft sensor model are determined, then according to the GRNN algorithm steps, a GRNN soft sensor model is established for biological parameters of the marine protease fermentation, and is compared by simulation with the RBF neural network soft sensor model. The results show that, in comparison with RBF neural network, the GRNN soft sensor model has the faster convergence speed, higher precision and stronger generalization ability.