Considered that key variables in glutamate fermentation process could not be measured inline

which would make it difficult to control and optimize the fermentation process

a model for glutamate concentration prediction in a 5 L fermentation tank was established on the basis of partial least squares (PLS) and least square support vector machine (LSSVM). PLS was applied first to extract features of input variables

to reduce number of variable dimensions

and to eliminate correlations such that model complexity was simplified and performance was improved. Coupled simulated annealing (CSA) arithmetic was later combined with grid search to determine model parameter values of LSSVM for improved prediction accuracy. Further model simplification was completed by deleting parameters with weak correlation to glutamate concentration. The simplified model was compared to kinetic model in order to select the best model of glutamate fermentation. Experimental results showed that the simplified LSSVM model equipped with CSA parameter optimization outperformed both PLS and kinetic models

which root mean square errors (RMSE) were 1.597

8.49 and 2.934 respectively. The LSSVM prediction model had excellent performance with high accuracy

so it would be more suitable for online prediction of glutamate concentration and offer an effective guidance for control and optimization of the glutamate fermentation process.