Aiming at the problem that crucial biological parameters are difficult to be measured online

and offline measuring cannot satisfy the needs of real-time optimized control due to the problem of big time-lag during the marine biological enzyme fermentation process

a soft measurement modeling method based on Bayesian regularization neural network is proposed. In this method

Bayesian regularization is used to limit the weights of network by modifying the neural network training property function

so the network structure is simplified

over fitting is restrained effectively and the generalization capability of neural network is enhanced. This method is applied to marine protease enzyme（ typical marine biological enzyme） fermentation process

measurement model for crucial biological parameters（ such as substrate concentration

biomass concentration and enzyme activity） for marine protease enzyme fermentation process is established. Simulation results show that Bayesian regularization neural network not only has strong capability of fitting for the training sample sets

and also has a strong generalization ability for prediction samples

the prediction precision is significantly higher than the LM algorithm neural network. The problem of on-line measurement of crucial biological parameters is solved for marine biological enzyme fermentation process.