Abstract：To solve the problems that the quality variables of fermentation products in the lysine production process were difficult to measure in real time and the existing soft measurement models had low accuracy and poor robustness, a multimodel soft measurement method was proposed based on an improved satisfactory clustering algorithm and the least squares support vector regression (ISCALSSVR) for lysine fermentation process. The sample data set was divided into c subsets by ISCA, and LSSVR machine was used to construct submodels for each subset separately. The particle swarm optimization algorithm and the annealing algorithm were used to collaboratively optimize the model parameters, and the output of each submodel was weighted and fused to obtain the final system output. The intelligent realtime monitoring system for key variables in lysine fermentation process was designed with the upper computer data processing module and the lower computer data acquisition module. The experimental simulation results show that compared with the traditional single LSSVR prediction model, by the ISCALSSVR model, the mass concentration prediction accuracies of product, substrate and cell are respectively increased by 501%, 362% and 678%, and the generalization ability is greatly improved.
